Part Number Hot Search : 
PE91034 7812CT 01V9A 3R3NZ ST78L18 M5YS0 SC2272 GL8T030
Product Description
Full Text Search
 

To Download MSM6650 Datasheet File
  If you can't view the Datasheet, Please click here to try to view without PDF Reader .  
　
 

  Datasheet File OCR Text:
  msm6679al-110 voice recognition processor first edition issue date: nov. 1998
msm6679al-110 voice recognition processor ? semiconductor contents general description ............................................................................................................ ............... 1 features ....................................................................................................................... ......................... 1 functional and i/o diagrams .................................................................................................... ...... 2 pin descriptions ............................................................................................................... .................. 6 electrical specifications ...................................................................................................... ............... 10 absolute maximum ratings ............................................................................................ 10 operating conditions ....................................................................................................... 10 dc characteristics (vdd = 2.7 to 5.5 v, ta = -30 to 70?c) .......................................... 11 ac characteristics ............................................................................................................ 1 2 timing diagram ................................................................................................................ 1 3 system configuration example ................................................................................................... .... 14 functional description ......................................................................................................... ............. 15 voice recognition ............................................................................................................. 1 5 si recognition ................................................................................................... 15 sd recognition ................................................................................................. 18 name tag recording ........................................................................................................ 19 audio input interface ....................................................................................................... 19 audio output interface .................................................................................................... 19 memory interface .............................................................................................................. 1 9 external voice synthesis control ................................................................................... 21 serial interface ............................................................................................................... .... 22 msm6679al-110 slave-mode api .................................................................................................. 2 3 command summary ........................................................................................................ 24 command descriptions ................................................................................................... 28 asynchronous serial protocol example ........................................................................ 40
? semiconductor msm6679al-110 voice recognition processor 1 ? semiconductor msm6679al-110 si/sd voice recognizer, recorder/player, and speech synthesizer general description the msm6679al-110 voice recognition processor (vrp) is a slave-mode device that performs five func-tions: speaker-independent (si) voice recognition, speaker-dependent (sd) voice recognition, solid-state sound recording, sound playback, and speech synthesis. the highly integrated device also provides an on-chip memory controller, flash memory interface, analog data conversion, oki speech synthesizer interface, and pulse width modulation (pwm) sound output. for si recognition, the msm6679al-110 contains a vocabulary template in external memory. pretrained si vocabularies eliminate the need for laborious training, as usually required by sd products. the memory requirements are dependent on the size of the vocabulary. the msm6679al-110 can tolerate background noise, while providing high recognition accuracy. in its designated operating environment, the device achieves a typical recognition accuracy of >95% (using an oki-defined test procedure). for sd recognition, the msm6679al-110 stores sd vocabulary templates, as defined by the user, in external sram. the msm6679al-110 can create sd vocabularies of up to 61 words each, with each word using approximately 50 bytes. in addition to providing voice recognition capabilities, the msm6679al-110 integrates a solid- state recorder/player, speech synthesis functions, and a tone generator. adpcm recording/ playback provides high quality sound and efficient memory utilization. the msm6679al-110 can respond to spoken com-mands, verbally or with tones, via an on-chip speech synthesizer and tone generator. for larger speech-synthesis requirements, the msm6679al-110 also provides a glueless msm665x control interface for off-chip speech synthesis. the msm6679al-110 can interface to any application or personal computer via a serial interface through an open, device-independent serial mode api (smapi). to accelerate code development, oki supplies an evaluation kit, and assembly and c language programs for this product. the msm6679al-110 is a low power version of the msm6679a-110. note: this device is intended for use in applications other than central office communication systems and central office switching systems. features this version: nov. 1998 e2f0013-28-y1 ? si recognition - up to 20 - 25 words in each vocabulary - multiple vocabulary support ? sd recognition - up to 61 words in each vocabulary - multiple vocabulary support ? speech synthesis - up to 2.3-sec internal and 27.6-sec external speech synthesis on-chip; sample looping and concatenation allows even longer phrases. - on-chip controller for msm665x speech synthesizer - standard beep tone outputs - pulse code modualation (pcm) and adaptive differential pulse code modualation (adpcm) voice or sound- effect output ? speech capture and playback - 28-kbps adpcm speech compression ? serial ascii command interface ? 6944-hz audio input sample rate for record and playback ? 10-khz sample rate for voice recognition ? 200-msec recognition latency ? flexible memory mapping for eprom, flash, and sram ? 14.3182 mhz operation ? package: 100-pin tqfp (tqfp100-p-1414-0.50-k)
msm6679al-110 voice recognition processor ? semiconductor 2 functional and i/o diagrams analog input recognition and synthesis engine vocabulary memory algorithm memory external memory control external speech synthesis control system controller serial interface pwm output figure 1. msm6679al-110 block diagram figure 2. msm6679al-110 logic symbol a/d interface adc0 ~ adc7 vref nar busy strobe si sd resout voiceout1 rxd txd res osc0 osc1 d0 ~ d7 wrram rdram rampage1 rampage0 pdc serial-mode msm665x interface pwm output serial interface ic reset and oscillator inputs memory interface a0 ~ a15 sleep
? semiconductor msm6679al-110 voice recognition processor 3 figure 3. msm6679al-110 100-pin tqfp pinout 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 wrram rdram n/c n/c sleep n/c n/c n/c n/c v dd osc1 osc0 gnd n/c n/c v dd n/c n/c res rampage1 rampage0 n/c n/c n/c n/c n/c n/c n/c n/c v dd vref adc0 adc1 adc2 adc3 adc4 adc5 adc6 adc7 agnd rxd txd gnd n/c n/c n/c n/c n/c n/c n/c n/c n/c n/c n/c n/c pdc n/c n/c n/c n/c n/c gnd v dd nar n/c voiceout1 n/c n/c sd si busy strobe resout n/c n/c 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 d0 d1 d2 d3 d4 d5 d6 d7 gnd a0 a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 a12 a13 a14 a15
msm6679al-110 voice recognition processor ? semiconductor 4 msm6679al-110 alphabetic pin list name # name # name # name # name # name # a0 60 a11 71 adc5 87 resout 3 vref 81 a1 61 a12 72 adc6 88 gnd 14,38, 59,93 wrram 50 a2 62 a13 73 adc7 89 a3 63 a14 74 a4 64 a15 75 nar 12 rxd 91 a5 65 agnd 90 osc0 39 sd 7 a6 66 adc0 82 osc1 40 a7 67 adc1 83 rampage0 30 a8 68 adc2 84 d4 55 rampage1 31 txd 92 a9 69 rdram 49 v dd 13, 35, 41, 80 a10 70 adc4 86 res 32 voiceout1 10 adc3 85 d0 51 d1 52 d2 53 d3 54 busy 5 d5 56 d6 57 d7 58 pdc 20 si 6 sleep 46 strobe 4
? semiconductor msm6679al-110 voice recognition processor 5 figure 4. msm6679al-110 100-pin package mechanical drawing
msm6679al-110 voice recognition processor ? semiconductor 6 pin descriptions pin # pin name signal type description 1 nc (do not connect) 2 reserved. these pins are reserved for future used and must be left open. 3 resout output msm665x reaet. this pin provides a reset signal for an external speech synthesis engine. 4 strobe output msm665x strobe. this output provides the load signal for an external speech synthesizer. 5 busy input msm665x busy. when using an external msm665x device, this pin monitors the msm665x busy signal and connects directly to the msm665x busy signal output. 6 si output msm665x serial clock. this msm6679al-110 output connects to the msm665x si input. the si pin is the msm665x serial clock input pin. 7 sd output msm665x serial data. this msm6679al-110 output connects to the msm665x sd input. the sd pin is the msm665x serial data input pin. 8 nc (do not connect) reserved. these pins are reserved for future use and must be left open. 9 10 voiceout1 output voice out. this pin is the pwm output for speech synthesis, voice sample playback, and voice prompts. an external integrator must be used to convert this to an analog signal. 11 nc (do not connect) reserved. this pin is reserved for future use and must be left open. 12 nar input msm665x next address request. this pin signals to the msm6679al-110 that the external speech synthesis engine is ready for another command. 13 v dd digital power power. 14 gnd digital ground ground. 15 nc input reserved. these pins are reserved for future use and must be tied to vdd. 16 17 18 nc (do not connect) reserved. this pin is reserved for future use and must be left open. 19 nc input reserved. this pin is reserved for future use and must be tied to vdd. 20 pdc input power down release. power down mode is released by both edge of pdc signal.
? semiconductor msm6679al-110 voice recognition processor 7 pin # pin name signal type description 30 rampage0 output ram page select. these signals support selection of one out of four ram pages. each page is 64kbytes in size. 31 32 res input msm6679al-110 reset. external logic should assert this power-on reset signal low when power is applied to the msm6679al-110. 35 v dd digital power 21 nc (do not connect) reserved. these pins are reserved for future use and must be left open. 22 23 24 25 26 27 28 29 rampage1 33 nc input reserved. these pins are reserved for future use and must be tied to vdd. 34 power. 38 gnd ground 36 nc input reserved. these pins are reserved for future use and must be tied to vdd. 37 ground. 39 osc0 input oscillator 0/external clock. when the msm6679al-110 uses a crystal oscillator, this input is the oscillator input pin. the pin is then connected to one side of a crystal and load capacitor. when used with an external clock, the external clock is applied to this input. 40 osc1 output oscillator 1. when the msm6679al-110 uses a crystal oscillator, this output is the oscillator output pin. the pin is then connected to one side of a crystal and load capacitor. when used with an external clock, this output is left unconnected. 41 v dd digital power power. 42 nc (do not connect) reserved. these pins are reserved for future use and must be left open. 43 44 45 46 sleep output sleep. when power down mode, this pin becomes low. sleep signal can be used for external memory control.
msm6679al-110 voice recognition processor ? semiconductor 8 pin # pin name signal type description 51 d0 bidirectional i/o memory data bus. 52 d1 59 gnd digital ground ground. 47 nc (do not connect) reserved. these pins are reserved for future use and must be left open. 48 49 rdram output ram read. this is a strobe signal for direct connection to an external ram's rd input. when asserted low, this signal indicates that the msm6679al- 110 is ready to read data from ram. 50 wrram output ram write. this is a strobe signal for direct connection to an external ram's wr input. when asserted low, this signal indicates that the msm6679al- 110 is ready to write data to ram. 53 d2 54 d3 55 d4 56 d5 57 d6 58 d7 60 a0 output memory address bus. 61 a1 62 a2 63 a3 64 a4 65 a5 66 a6 67 a7 68 a8 69 a9 70 a10 71 a11 72 a12 73 a13 74 a14 75 a15
? semiconductor msm6679al-110 voice recognition processor 9 pin # pin name signal type description 82 adc0 analog input analog input. these eight inputs are tied together and serve as the analog input. signal conditioning, via a bandpass fillter and gain circuit, is required before this input. 83 76 nc (do not connect) reserved. these pins are reserved for future use and must be left open. 77 78 79 adc1 90 agnd analog ground analog ground. this pin provides an analog ground point, allowing independent grounding of the analog and digital circuitry. separate grounds reduce the impact of digital switching noise on analog sampling accuracy. 80 v dd digital power power. 81 vref analog power reference voltage analog power. the msm6679al-110's on-chip a/d converter uses this analog power when converting an analog signal into digital samples. also this is used as an analog reference voltage. 84 adc2 85 adc3 86 adc4 87 adc5 88 adc6 89 adc7 91 rxd input serial port receive. this is the receive data line for serial port. 92 txd output serial port transmit. this is the transmit data line for serial port. 93 gnd ground ground. 94 nc (do not connect) reserved. these pins are reserved for future use and must be left open. 95 96 97 98 99 100
msm6679al-110 voice recognition processor ? semiconductor 10 electrical specifications absolute maximum ratings 1. permanent device damage may occur if absolute maximum ratings are exceeded. functional operation should be restricted to the conditions as detailed elsewhere in this data sheet. exposure to absolute maximum rating conditions for extended periods may affect device reliability. operating conditions digital power supply voltage v dd symbol conditions C0.3 to +7.0 value parameter input voltage v i C0.3 to v dd +0.3 output voltage v o gnd = agnd = 0 v C0.3 to v dd +0.3 v analog power/reference voltage v ref C0.3 to v dd +0.3 analog input voltage v ai C0.3 to v ref power dissipation pd ta = 70?c, per package 650 mw ta = 70?c, per output 8 storage temperature t stg C50 to +150?c ?c unit digital power supply voltage v dd symbol f osc = 14.3182 mhz conditions 2.7 to 5.5 value parameter analog power/reference voltage v ref v dd C0.3 to v dd v analog input voltage v ai a gnd to v ref storage holding voltage v ddh f osc = 0 mhz 2.0 to 5.5 operating frequency f osc v dd = 2.7 to 5.5 v 14.3182 mhz ambient temperature ta C30 to 70?c ?c mos load 20 fan-out n ttl load, d0 ~ d7, wrram , rdram and sleep 6 ttl load, all other outputs 1 unit
? semiconductor msm6679al-110 voice recognition processor 11 dc characteristics (vdd = 2.7 to 5.5 v, ta = -30 to 70?c) 1. typical condition is 3 v 25?c. parameter symbol condition min typ [1] rated value max unit 0.44 v dd v dd +0.3 applied to d0-d7 high-level input voltage v ih 0.80 v dd v dd +0.3 applied to all other i/o C0.3 0.16 v dd applied to d0-d7 low-level input voltage v il C0.3 0.2 v dd applied to all other i/o high-level output voltage v oh v dd C0.4 v output current = C400 m a, applied to d0-d7, wrram , rdram and sleep v dd C0.4 output current = C200 m a, for all other i/o low-level output voltage v ol 0.5 output current = 3.2 ma, applied to d0-d7, wrram , rdram and sleep 0.5 output current = 1.6 ma, for all other i/o input leak current i ih , i il m a 1/C1 v i = v dd /0 v, applied to adc0- adc7 input current 1/C250 v i = v dd /0 v, applied to res 15/C15 v i = v dd /0 v, applied to osc0 high-level output current i oh C2 v o = 2.4 v, applied to d0-d7 C1 ma v o = 2.4 v, applied to all other i/o low-level output current i ol 10 v o = 2.4 v, applied to d0-d7 5 v o = 2.4 v, applied to all other i/o output leakage current i lo m a 10 v o = v dd /0 v input capacitance c i pf 5 f = 1 mhz, ta = 25?c output capacitance c o 7 analog reference power supply voltage i ref ma 4 during voice input m a 10 when voice input is halted power consumption i dd ma t.b.d f osc = 14.3182 mhz, no load
msm6679al-110 voice recognition processor ? semiconductor 12 ac characteristics external data memory control (vdd = 2.7 ~ 5.5 v, ta = -30 ~ 70?c) parameter symbol condition min. max. unit cycle time t cyc 69.8 clock pulse width (high level) t f wh 28 rdram pulse width t rw 190 wrram pulse width t ww 190 rdram pulse delay time t rd 75 wrram pulse delay time t wd 75 address set-up time t as C5.1 address hold time t ah 29 ns 41 c l = 50 pf read data set-up time t rs 60 read data hold time t rh 0 read data access time t acc 124 write data set-up time t ws 169 write data hold time t wh 29 41 clock pulse width (low level) t f wl 28
? semiconductor msm6679al-110 voice recognition processor 13 timing diagram t cyc t f wh t f wl clk rdram t rw t rd a0 - a15 d0 - d7 din0 - 7 wrram t ww t wd rap0 - 15 t ah t as t acc t rs t rh a0 - a15 d0 - d7 dout0 - 7 rap0 - 15 t ah t as t wh t ws clk : clock pulse wrram : ram write strobe signal rdram : ram read strobe signal a0 - a15 : memory address bus rap0 - 15 : ram address din0 - 7 : read data dout0 - 7 : write data figure 5. ram read/write timing
msm6679al-110 voice recognition processor ? semiconductor 14 system configuration example msm6679al-110 sram flash d0 d1 d2 d3 d4 d5 d6 a0 a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 a12 a13 a14 a15 sleep wr rd rampage0 rampage1 osc0 osc1 d7 d2 d3 d4 d5 d6 d7 d1 d0 a2 a3 a4 a5 a6 a7 a1 a0 a8 wr a9 rd a10 cs a11 a12 a13 a14 d2 d3 d4 d5 d6 d7 d1 d0 a2 a3 a4 a5 a6 a7 a1 a0 a8 wr a9 rd a10 cs a11 a12 a13 a14 a15 a16 3 2 4 33 14.3182 mhz 0.1 m f 0.1 m f 0.1 m f txd rxd pdc adc0 adc1 adc2 adc3 adc4 adc5 adc6 adc7 vref agnd voiceout analog circuit msm66p54 busy si sd st reset nar aout xt nar busy si sd strobe resout xt v dd v ss 33 0.22 m f 22 m f mic speaker host mcu interface 4 mhz v dd gnd figure 6. msm6679al-110 system configuration example
? semiconductor msm6679al-110 voice recognition processor 15 functional description voice recognition the msm6679al-110 performs both si and sd recognition. si vocabularies are embedded in the msm6679al-110. for sd recognition, each recognized phrase must be enrolled in the msm6679al-110s vocabulary by creating a composite template from multiple recordings of the same phrase. then the composite tempalte is stored in sram or flash memory. during both si and sd recognition, the msm6679al-110 performs the following steps: 1. after external band-pass filtering, the msm6679al-110 converts the analog signal to pcm samples. 2. the msm6679al-110 extracts significant features from the sample data by frequency and time-domain analysis. 3. the msm6679al-110 compares the analyzed input with the reference data for each signal, weighing the significance of similarities according to control software parameters. a score (expressed as distance) is generated for each phrase. 4. the vocabulary phrase that achieves the highest score (or lowest distance) is judged to match the input phrase, assuming that the score exceeds a predetermined threshold. 5. via a special command, the msm6679al-110 can also return the scores of the input against all defined vocabulary phrases for si or sd recognition. this feature allows external host software to select the next best match, if the closest match is not contextually logical. si recognition oki supplies the msm6679al-110 with predefined si vocabularies which oki builds from hundreds of utterances by a wide variety of speakers. si vocabularies are limited to 25 words or less, which allows the msm6679al-110 to achieve a net accuracy of >95%, even in noisy conditions. si vocabularies are grouped into sub-vocabularies of 15 words, to maintain the highest accuracy. similar words in any one sub-vocabulary can cause substitution errors. oki semiconductors standard cellular vocabulary is intended for an automotive environment with a far-talk microphone. this vocabulary may work adequately in other conditions, such as an office or outside, but recognition performance may be degraded. msm6679al-110 cellular si recognition vocabulary sub-vocabulary 1 sub-vocabulary 2 sub-vocabulary 3 phrase store index phrase index phrase index phrase index 1 one 1 yes 1 dial 2 two 2 eight 8 no 2 delete 3 three 3 nine 9 cancel 3 directory 4 four 4 zero ah five 5 oh bh six 6 stop ch seven 7 clear dh
msm6679al-110 voice recognition processor ? semiconductor 16 msm6679al-110 control vocabulary sub-vocabulary 1 sub-vocabulary 2 phrase a/c index phrase index 1 low 1 fan 2 medium 2 temperature 3 high 3 timer 4 increase 4 service 5 decresse 5 help 6 set 6 select 7 reset 7 cancel 8 clear 9 recall a on b help c msm6679al-110 direction vocabulary sub-vocabulary 1 phrase up index 1 down 2 left 3 right 4 formard 5 reverse 6 faster 7 slower 8 start 9 stop a cancel b msm6679al-110 browse vocabulary sub-vocabulary 1 sub-vocabulary 2 phrase up index phrase index phrase index phrase index phrase index 1 next 5 home 9 set 1 on 5 down 2 previous 6 reset 2 play 6 left 3 select 7 start 3 lock 7 right 4 cancel 8 stop 4 cancel 8
? semiconductor msm6679al-110 voice recognition processor 17 msm6679al-110 japanese navigation vocabulary sub-vocabulary 1 sub-vocabulary 2 sub-vocabulary 3 sub-vocabulary 4 phrase genzaichi index phrase index phrase index phrase index 1 ue 1 hyoujun 1 hai 1 jiaku 2 shita 2 kakudai 2 iie 2 kaisya 3 hidari 3 shukushou 3 ofu 3 houi 4 migi 4 zentai 4 sentaku 5 kaiten 5 yuudou 6 kyori 6 nabi 7 hosei 7 teisei 8 msm6679al-110 japanese cellular vocabulary sub-vocabulary 1 sub-vocabulary 2 phrase on index phrase index phrase index 1 ichi 1 kyuu 9 ofu 2 ni 2 zero a daiyaru 3 san 3 sharp b tansyuku 4 yon 4 star c denwacho 5 go 5 kakunin d kakunin 6 roku 6 touroku e nabi 7 nana 7 rei f hachi 8 msm6679al-110 german cellular vocabulary sub-vocabulary 1 sub-vocabulary 2 phrase speichern index phrase index phrase index 1 eins 1 neun 9 w?hlen 2 zwei 2 null a l?schen 3 drei 3 notruf b name 4 vier 4 w?hlen c fnf 5 l?schen d sechs 6 raute e sieben 7 stern f acht 8 sub-vocabulary 3 phrase ja index 1 nein 2 l?schen 3 si vocabulary generation starts with collecting reference utterances from 3 400 speakers with: ? an equal mixture of males and females ? accents from all regions of the country of intended use ? ~15% non-native speakers.
msm6679al-110 voice recognition processor ? semiconductor 18 the samples should be generated from a randomly-ordered list, with each word spoken twice and with a dummy word at the beginning and end. there must be >2 sec between each sample for accurate data processing. to provide the audio fidelity required for high-quality recognition training, a dat recorder, together with the microphone that will be used in the final application, is required. to ensure data integrity, data is submitted to oki after collecting samples from the first 20 speakers for initial screening. if acceptable, then the remaining collection may proceed. if substitution errors are possible, collection of spare words during initial collection is recommended. for example, alternate words to stop and top could be halt and first. collections should contain a wide variety of the background sound conditions that will exist during actual usage. for example, if the collection is for use in an automobile, conditions such as vehicle speed, road conditions, various window opening positions, heater or ac blower speeds and radio volumes should be varied during the collection. the signal-to-noise ratio should be maintained at 3 20db. to achieve high accuracy rates, phrase selection, data collection, background initialization strategy, and control software need careful consideration. there are no published standards for recognition accuracy. oki defines accuracy by: accuracy = 100% - e rate e rate = e sub + 1/2 e rej with the following definitions: parameters for recognition accuracy a typical target accuracy of 97% is achieved with a 3% e rate , composed of a 1.5% e sub rate and a 3%e rej rate. sd recognition in sd recognition mode, the msm6679al-110 can be trained to recognize up to 61 words. the msm6679al-110 can support multiple speakers by switching vocabularies, but only one speakers vocabulary should be active at one time. the end user enrolls a phrase in the msm6679al-110s vocabulary by recording the phrase three times or more. the host micro controller unit (mcu) controls the number of times each phrase in enrolled. generally, higher recognition accuracy is achieved with each additional enrollment. the word set is made more robust by pronouncing each phrase slightly differently during initial enrollment. in addition to enrollment training, adaptive template updating can drive the accuracy towards 100%. the host mcu updates templates by first asking the speaker to confirm a recognized phrase with a yes or no response, and subsequently updating the template for corresponding words. the use of name tags (see next paragraph) facilitates this process. name condition symbol substitution error most critical type error, e.g., say "five", recogrize "nine" e sub rejection error word not recognized, opportunity for operator to repeat e rej gap error word spoken before recognizer ready e gap time-out error word length is too long e tme spurious response error sourd or imvalid word classfied as a valid word (i.e., drop handset or speak wong word) e spu
? semiconductor msm6679al-110 voice recognition processor 19 name tag recording to facilitate sd recognition, the msm6679al-110 supports recording and playback of name tags. name tags are used to confirm correct responses in sd recognition. for example, in a phone dialer application, the user associates a name (which is recorded into memory) with a phone number. the msm6679al-110 then plays back the name tag so that the user can verify that the recognized phrase is the correct one. the vrp stores names tags in memory using an adpcm compression algorithm with 28 kbps of speech. the length of a name tag is controlled with a command from the users host mcu program. the maximum number of name tags possible is 61, but the actual number is dependent upon record time and memory available. see the section on memory interface for more detail. audio input interface a critical item for high-accuracy speech recognition is correct design of the audio input circuit. a circuit with appropriate gain and frequency responses must be placed between the microphone and msm6679al-110s a/d input. oki recommends input gain and a band pass filter with the following characteristics: ? four pole chebyshev high-pass filter, 3 db point at 225 hz ? dual-pole low-pass filter, 3 db point at 4250 hz ? midband gain of 46 db at 1000 hz the above gain and filter characteristics are obtained by using a rail-to-rail quad cmos op-amp and one-half supply rail splitter to bias the input signal at 1/2 v dd nominal. the msm6679al-110 uses multiple analog inputs to improve sampling quality. an on-chip analogy to digital (a/d) conversion unit transforms the analog signal to a digital data stream. audio output interface the msm6679al-110 also provides the voiceout1 pwm output. the msm6679al-110 uses adpcm to generate voice or sound-effect output. adpcm represents an improvement over conventional pcm techniques in that it adaptively changes the quantizer step (scale factor) to suit the waveform being encoded. the result is more efficient memory usage with no loss of quality. careful selection of the components for internal and external output filters and amplifiers is recommended. an incorrect choice would impair the original quality. this consideration equally includes: ? careful separation of analog and digital lines ? grounding of analog lines at both ends ? further adequate separation from high-speed digital circuits to avoid distortions thereof memory interface the memory control section manages ram and/or rom devices in two 64-kbyte memory spaces, in conjunction with internal memory for voice templates and working memory. some versions work with no external memory, some have some external ram, some use only external eprom, and some use external memory in conjunction with both internal rom and ram. the msm6679al-110 requires a minimum of 32 kbytes sram and 16 kbytes rom. the following table shows vocabulary sizes and playback facilities for various configurations.
msm6679al-110 voice recognition processor ? semiconductor 20 typical configurations 1. phrase chaining features usually permit much longer overall playback durations; not including external speech synthesizer. 2. sd recognition vocabularies are volatile in these configurations. 3. per download. vocabulary swapping by host permits unlimited vocabulary size. the msm6679al-110 supports 32 kbytes of ram, and up to 64 kbytes of rom (eprom or flash) per bank in separate memory spaces. for accessing the rom and ram address spaces, the msm6679al-110 provides the separate write ram ( wrram ) and read ram ( rdram ) signals. the rdram signals connect directly to output enable ( oe ) control signal inputs on the ram and rom, respectively. the wrram signal connects directly to the write enable ( we ) control signal input on the ram. application recognition vocabulary (words) msm6679al-110 sound playback (sec) [1] msm665x playback interface msm6679al-110 speech record msm6679al-110 speech playback memory size (bytes) si sd internal external eprom flash sram 25 61 [2] 2.3 9.2 ok ok controller 50 61 [2] 2.3 ok ok 64k 32k 25 61 2.3 27.6 ok ok ok 50 61 2.3 18.4 ok ok ok 128k 32k telephone dialer 75 61 2.3 ok ok ok 100 61 2.3 ok ok ok 61 [3] 61 2.3 36.8 ok ok 64-384k computer peripheral 12 61 [2] 1.15 ok 16k 32k minimum configuration
? semiconductor msm6679al-110 voice recognition processor 21 si first (f509*) flash 00000 000 name tag block address 07300 07d80 08000 10000 18000 si last (f501*) sd first 05480 ntp first name tag data 1f300 sd last 1fd80 ntp last 1ffff *denotes commands to select blocks sram reserved alternate sd templates default working sd templates working name tag pointer table 04a00 05700 08000 100 200 2f6 2fb 2ff figure 7. msm6679al-110 external memory map external voice synthesis control the msm6679al-110 is capable of interfacing to the msm665x family of oki rom, otp, or external eprom speech synthesizers, allowing for up to 260 seconds of high-quality voice and sound effects. the following table indicates the speech capabilities of the msm665x family. msm665x family characteristics type MSM6650 data rom capacity [1] 64 mbits [3] >1 hour >40 minutes maximum speech duration [2] >30 minutes >15 minutes f sam = 4.0 khz f sam = 6.4 khz f sam = 8.0 khz f sam = 16.0 khz f sam = 32.0 khz >8 minutes msm6652 288 kbit 16.9 sec 10.5 sec 8.4 sec 4.2 sec 2.1 sec msm6653 544 kbit 31.2 sec 19.5 sec 15.6 sec 7.8 sec 3.9 sec msm66p54 [4] 1 mbit 63.8 sec 39.9 sec 31.9 sec 15.9 sec 7.9 sec msm6654 1 mbit 63.8 sec 39.9 sec 31.9 sec 15.9 sec 7.9 sec msm6655 1.5 mbit 96.5 sec 60.3 sec 48.2 sec 24.1 sec 12.0 sec msm66p56 [5] 2 mbit 129.1 sec 80.7 sec 64.5 sec 32.2 sec 16.1 sec msm6656 2 mbit 129.1 sec 80.7 sec 64.5 sec 32.2 sec 16.1 sec msm6658 4 mbit 258 sec 161.4 sec 129.1 sec 64.5 sec 32.2 sec 1. actual rom area in msm6652, msm6653, msm6654, msm6655, and msm6656, msm6658, msm66p54, msm66p56 is smaller by 22 kbits.
msm6679al-110 voice recognition processor ? semiconductor 22 2. longer speech patterns can be created by chaining and repeating existing speech samples. 3. via external rom only (no on-chip rom available). 4. one-time-programmable (otp) version of msm6654. see the msm66p54 data sheet for more information. 5. one-time-programmable (otp) version of msm6656. see the msm66p56 data sheet for more information. the msm665x interface consists of the following signals: ? busy - asserted low during msm665x device playback. the msm6679al-110 f50bh and f10100xxh commands select this signal for msm665x command polling. ? nar - next address request status signal. by default, the msm6679al-110 uses this signal to poll commands to the msm665x. the f51bh, f480h, and f440h commands select nar for polling. ? si - serial input clock. ? sd - serial data out. ? strobe - initiates speech synthesis. ? resout - initializes device when asserted low. the msm6679al-110 f480h command generates this signal. serial interface the msm6679al-110 supplies a serial interface suitable for connection to an rs-232c serial port buffer or equivalent. the serial interface uses one msm6679al-110 input (rxd) and one msm6679al-110 output (txd). the interface operates at 9600 baud with: ? 8 data bits ? 1 start bit ? 1 stop bit ? no parity ? no handshake a host processor sends serial ascii commands to the msm6679al-110 and receives serial ascii responses based on voice input responses.
? semiconductor msm6679al-110 voice recognition processor 23 msm6679al-110 slave-mode api this section describes the slave-mode applications protocol interface (api) between a host mcu and the msm6679al-110. the slave-mode api offers the following features: ? direct slave-mode control voice recognition, sound recording and playback, and sound synthesis ? serial port interfaces ? simple procedures for downloading and uploading data ? ascii format ? comprehensive return codes and error reporting the host mcu selects the active speech recognition vocabulary, speech responses, and controls all actions required to implement an interactive voice response system. the msm6679al-110 performs speech recognition, based on the vocabulary selected by the host, and returns digital codes representing the most probable match of the current utterance to an individual utterance in the selected vocabulary. the msm6679al-110 can also respond with name tags. name tags can be fixed words, phrases or sound effects, or can be words, phrases or sound effects that have been interactively recorded by the user. the api supports serial interface. the msm6679al-110 returns each response using the same interface through which the most recent message was received. the user can thus connect and use both interfaces. for all messages, the serial interface represents each 8-bit value with two hexadecimal digits coded in ascii. when downloading and uploading data, the msm6679al-110 uses a stream of 8-bit binary values. the serial-mode interface uses a 9600-baud uart with 1 start bit, 8 data bits, and 1 stop bit. there is no parity or handshaking. serial-interface messages are of variable length, but consist of an even number of bytes. the serial interface echoes all received ascii characters immediately back to the host mcu. messages are of variable length. all messages consist of an even number of bytes. opcodes consist of exactly four bytes, with values between f000h and fefeh. operand bytes may take values from 0000h to ffffh. the msm6679al-110 issues a return code for many of the host commands. the return code generally consists of the same opcode, followed by data indicating success of failure of the operation. opcodes are organized into the following categories: ? purge ? set parameter ? initialize ? recognize ? speak ? request ? record ? sd recognition control the following tables summarize available opcodes and provide detailed descriptions of the opcode functions.
msm6679al-110 voice recognition processor ? semiconductor 24 command summary function purge opcode (hex) description f000 clear msm6679al-110 input stack default (hex) set parameter f102 xxxx f103 xxxx f104 xxxx f12x f130 xxxx set sp/si origin to xxxx. set sd origin. set triggering origin. set sd sp table to table x. select triggering table. 8000 4a00 f100 f123 0101, 0202... initialize f2xx mod 80 f2xx mod 40 f2xx mod 20 f2xx mod 10 f2xx mod 8 f2xx mod 4 f2xx mod 2 f2xx mod 1 initialize background estimation. wait for f3h command after each response. beep after each triggered utterance reserved set speech response level to default. send acknowledge after each speech output response. only detect triggers. initialize sd parameter table and name tags. disabled. disabled. disabled. disabled. enabled. enabled. disabled. load from first flash. recognize f300 f301 to f33f f340 f341 f342 f343 f344 f351 f361 f371 stop listening (recognition). start si recognition. start sd recognition. sort sd recognition distances, return index to utterance with least distance. update sd enrollment. request recognition parameter upload to host. sort sd recognition distances, return index and distance to utterance with least distance sort sd recognition distances, return all distances. sort sd recognition distances, return minimum and maximum energy values. sort sd recognition distances, return all energy values and distances. speak f401 to f43d f441 to f47c f47e f47f f480 f481 - f4ff f50b f51b fe03 to fefe play back name tag from external memory. play back sound from internal memory. play 50-ms beep. pause for 0.2 sec. initialize msm665x ic, set msm665x busy mode off, select flash si recognition. play back one of 127 phrases in external msm665x device. set msm665x busy mode on. set 6654 nar mode set output volume (03h = minimum, feh = maximum). off on fe80h request f500 f501 f510 f520 status request. select last flash bank for si recognition. select download ram bank for speaker independent/signal processing (si/sp) template area. set msm6679al-110 power down mode. f509 f509
? semiconductor msm6679al-110 voice recognition processor 25 function request opcode (hex) description f502.... f504 f505 f506 f507 f517 f508 f518 f509 download/upload. retrieve msm6679al-110 firmware revision. initialize background (bg) noise level. retrieve vocabulary and trigger table revision number. save sd templates from download ram to first flash. save sdr templates in last flash. (4a00-547b ? f300-fd7f) recall sd templates from first flash to download ram. get sdr templates from last flash (f300-fd7b ? 4a00-547b) select first flash bank for si recognition. default (hex) 414c 3039 f509 record f101 00xx f105 f106 f50a f50c f51c f50d f51d f50e f50f fa01 ~ fa3d set name tag length, set msm665x busy mode on. set name tag record origin set name tag record end clear name tag table in sram (5480 - 56ff). recall last saved name tag table. recall name tag pointers from last flash (fd80-ffff ? 5480-56ff) save name tag table from sram to flash. save name tag pointers in last flash (5480-56ff ? fd80-ffff) set record volume high. set record volume normal (default). record name tag 01h - 3dh. 0051 0000 01ff f50f f50f sd recognition control f6xx f9xx fb00 fc00 f521 set sd pointer to segment xxh. search for sd utterance xxh. enroll sd utterance selected by search command (f9xx). erase utterance from sd vocabulary. clear sdr table (4a00 - 547b) response summary command result after parameter set operands description f101h 00 tm f102h adh adl f103h adh adl f104h adh adl f12xh record time = tm*14 msec. high and low bytes of sp/si origin address. high and low bytes of sd origin address. high and low bytes of triggering origin address. sp table xh selected. initialization acknowledgment f280h f240h f220h f210h f208h f204h f202h f201h invalid message received. sample data over-run. [1] 32-kbyte block boundary violation error. unclassified download/upload error. divide-by-zero error. select/jump error. invalid sp header or table. reserved. speech ack f400h speech acknowledgment. [2]
msm6679al-110 voice recognition processor ? semiconductor 26 response summary (continued) command si recognition result [5] operands description f600h f6utt f6 utt dst1h dst1l...dstnh dstnl f6 utt eminh eminl emaxh emaxl f6 utt dst1h dst1l...dstnh dstnl eminh eminl emaxh emaxl f63ah f63bh f63ch f63dh f63eh f63fh aborting si listen mode. utt = utterance id. utterance id, high/low byte of distance to utterance 1...utterance n. utterance id, high/low byte of min. and max. energy value, utterance id, high/low byte of distance to utterance 1...utterance n, high/low byte of minimum energy value, high/low byte of maximum energy value. trigger detection code (see init command). rejection: utterance too loud. rejection: utterance too long. rejection: utterance begins too soon. rejection: bad signal/noise ratio. rejection: reason uncertain. sd recognition result f700h f73eh f73fh f740h f341h f7utt f344h f7utt dsth dstl f351h f7utt dst1h dst1l... dstnh dstnl f361h f7utt eminh eminl emaxh emaxl f371h f7utt dst1h dst1l... dstnh dstnl eminh eminl emaxh emaxl aborting sd listen mode. after sd utterance search: not found. rejection. sort completed. after sd utterance search: empty. rejection: msm6679al-110 sd memory full/empty. after sd utterance search: in use. utt = utterance id triggered. utterance id, high/low byte of distance. utterance id, high/low byte of distance to utterance 1... utterance n. utterance id, high/low byte of minimum energy value, maximum energy value. utterance id, high and low byte of distance to utterance 1... distance to utterance n, high and low byte of minimum energy value, maximum energy value. vector upload f743h 0000h f743h nh nl v1h v1l...vnh vnl upload failure. high/low bytes of length of vector, v, high/low byte of first v...nth v. trap error codes f801h f802h f804h f808h f810h f820h f840h f880h reserved. invalid sp header or table. select/jump error. divide-by-zero error. unclassified download/upload error. memory full; 32-kbyte block boundary violation error. sample data over-run. [1] invalid message received. record response fa00 record complete. status [3] f500h f501h f520h f540h f560h f580h f5a0h f5c0h f5f0h msm6679al-110 ready. operation complete. operations complete; msm6679al-110 disabled (vocabulary 0). msm6679al-110 waiting for start command. msm6679al-110 waiting for end trigger. msm6679al-110 processing recognition. download/upload in progress. [4] download/upload complete. speak output in progress.
? semiconductor msm6679al-110 voice recognition processor 27 1. sample data overrun issued when real-time sp in listen mode cannot keep up with incoming samples, i.e., if the a/d signal input routine overwrites a sample data buffer before it is fully processed. 2. this acknowledge is sent only if init command 1111 0010 xxxx x1xx (f2 xxxx x1xx) is set to enable acknowledgments. 3. these messages are sent in response to a request command (f5xyh) from the host. 4. upload/download in progress, acknowledging load request immediately before data transfer. if in response to an n-byte download request, the msm6679al-110 then receives n bytes (if n is even, or n+1 if n is odd) of data from the host. if n is odd and n+1 bytes are received, only n bytes are written to msm6679al-110 memory. if in response to an upload, the msm6679al-110 then sends n bytes (if n is even, or n+1 if n is odd) of data to the host. 5. if an utterance was recognized, xyh is the utterance identity or class number, and additional parameters may be appended, if requested in the si recog (f3xyh with x=0...3) command. otherwise, xyh indicates various results as detailed.
msm6679al-110 voice recognition processor ? semiconductor 28 command descriptions purge set parameter operand f000 return values description none purge msm6679al-110 input stack. this command clears the msm6679al-110 input stack of commands that are waiting to be executed. commands already in progress, such as a pending msm6654 poll action, are not affected. it does not affect the msm6679al-110 output stack. operand f102h xxyyh return values [1] description f102h xxyyh = high (xxh) and low (yyh) bytes of resultant address. if a valid header is not found at the resultant address, the msm6679al-110 immediately sends response code: f802h = invalid sp/si header. f103h xxyyh f103h xxyyh = high (xxh) and low (yyh) of resultant address. f104h xxyyh f104h xxyyh = high (xxh) and low (yyh) bytes of resultant address. set sp/si recognition origin. prior to sd or si recognition, address pointers must be set to point at the sp or si recognition parameter tables.this command sets the starting address of sp and si recognition parameter tables. this address is the location of the first word of a header that contains pointers to one or more individual sp/si tables. xxyyh = high (xxh) and low (yyh) bytes of requested address. the msm6679al-110 uses and returns an even address outside the msm6679al-110 work space that is as near as possible to the requested address. leave this parameter at its default value unless you are using an oki custom si vocabulary and are instructed to alter sp/si recognition origin. default sp/si origin: 8000h set sd recognition origin [2] . this command sets the sd origin address at the starting address of the current sd recognition parameter table. this command may be used to select among mul-tiple ram-resident sd vocabulary tables. xxyyh = high (xxh) and low (yyh) bytes of requested address. the msm6679al-110 uses and returns an even address outside the msm6679al-110 work space that is as near as possible to the requested address. leave this parameter at its default value unless you are using an oki custom vocabulary and are instructed to alter sd recognition origin. the table length is 0a7ch bytes. default sd origin: 4a00h set triggering origin. this command sets the starting address of triggering parameter tables. this address is the location of the first word of a section of data memory containing one or more contiguous triggering parameter tables. xxyyh = high (xxh) and low (yyh) bytes of requested address. the msm6679al-110 uses and returns an even address outside the msm6679al-110 work space that is as near as possible to the requested address. leave this parameter at its default value unless you are using an oki custom si vocabulary and are instructed to alter triggering origin. default triggering origin: f100h.
? semiconductor msm6679al-110 voice recognition processor 29 set parameter (continued) operand return values [1] description f12yh f12z = sp table z selected. if the sp header is invalid, a second message follows: f802h = invalid sp header. f130h vn tn f130h f(vn) f(tn) = triggering table selected. default = 0101, 0202, 0303... set sd recognition sp table. this command sets the sp parameter table number to be used in processing speech input during sd recognition. the msm6679al-110 selects sp table number z, where z is the nearest valid value to y. by default, the msm6679al-110 selects sp table 3 until this command is issued. this command selects sp parameters only, and does not select among multiple ram-resident sd vocabulary tables, which can be independently selected by the set sd origin command (f103h). after setting the table number and returning the resultant value, the msm6679al-110 checks the validity of the sp header. if the header is invalid, an error message is returned. set this value to (nsi +1), where nsi is the number of si subvocabularies. default sp table: 3. select triggering table. this command selects triggering table tn for use with sp table vn. valid values for vn and tn are between 01h and 0fh. leave this parameter at its default value unless you are using an oki custom si vocabulary and are instructed to alter the triggering table. 1. return value is actual parameter value which may not equal the set parameter value. 2. see also f6xy
msm6679al-110 voice recognition processor ? semiconductor 30 initialize f2xx bit values power-on/ reset value action return value 1xxx xxxx cleared f501 = background initialization complete f2xy = initialization acknowledge. [1] x1xx xxxx cleared f2xy = initialization acknowledge. [1] after power-on, the msm6679al-110's mode corresponds to that after issuing a f20c command. this mode may not be the optimum condition for most situations, so the user is advised to carefully understand the desired condition and develop a suitable command for the application at hand. in addition, ensure that unwanted bits do not get set or reset when attempting to set individual conditions. the conditions selected are based on the xxh values associated with the last f2 command issued. background noise initialization. when set to 1, the msm6679al- 110 starts a 500-ms background noise initialization. when set to 0, the msm6679al-110 does not perform background noise initialization. the msm6679al-110 requires this command prior to recognition for noise vector subtraction during the utterance sampling period. use the background initialization command whenever there is a change in the background noise level. for example, sample the noise signature in a vehicle at rest and moving at 35 mph with its windows rolled down. the quality of a phone line connection can also vary from call to call. the host mcu must implement a strategy as to when to issue a background initialization command. in a vehicle, the host mcu could monitor the vehicle speed, fan speed, radio volume, etc. alternatively, the host mcu could issue this command each time a new recognition session starts or a new line connection is established. however, the 0.5-sec sample period could degrade system responsiveness if used too frequently. a zero in this bit location during the f2xxh command will not cause an initialization. the f505h command causes the same initialization sequence. wait for recognition command/auto restart si recognition. when set to 1, the msm6679al-110 waits for a recognition command after each response. when set to 0, the msm6679al- 110 auto-restarts si recogni-tion after each response. this bit should be set to 1 when an action is to be taken immediately after an utterance. auto-restart recognition is the desired mode during digit string recognition, automated tape testing of digits, or in demonstrations where continuous recognition is desired.
? semiconductor msm6679al-110 voice recognition processor 31 initialize (continued) 1. see the response summary table earlier in this section for a complete description of the xy codes in initialization acknowledgment messages. f2xx bit values power-on/ reset value action return value xx1x xxxx cleared f2xy = initialization acknowledge. [1] xxxx 1xxx set f2xy = initialization acknowledge. [1] xxxx x1xx set f2xy = initialization acknowledge. [1] xxxx xx1x cleared f2xy = initialization acknowledge. [1] beep after each voice trigger. when set to 1, the msm6679al- 110 beeps after each voice trigger. when set to 0, the msm6679al- 110 does not beep after each voice trigger. these beeps do not cause a f400h message to be issued to the host mcu. when set to 1, the msm6679al-110 beep can help a user avoid speaking before the msm6679al-110 is ready. this mode is normally used with a digits vocabulary to pace the user and confirm each utterance reception. instead of using beeps, an external msm665x speech synthesizer can repeat digits as they are recognized. however, some users find the number repetition annoying. therefore, firmware could repeat digits during initial usage and switch to beep mode later. typically, performance improves with time as users learns to speak with the correct enunciation and volumes. the msm6679al-110 in this case trains the user. note that the host mcu can also make the msm6679al-110 beep with the f47eh command. set output volume. when set to 1, voiceout1 sound output level is set to half of full volume (80h). when set to 0, voice output level is unaffected. msm6679al-110 sound output volume can also be set at any level on a continuous scale from 00h to feh (low to high) with the fexxh command. the msm665x speech synthesizer has four discrete sound output volumes, corresponding to 0h - 20h, 21h - 40h, 41h - 80h, and 81h - feh. send response code after sound output. when set to 1, the msm6679al-110 issues an acknowledge response (f400h) when sound output is completed. when set to 0, the msm6679al-110 does not issue an acknowledge response when speech response is completed. automatic beeps after voice triggers do not cause an f400h command to be issued. trigger detection only. when set to 1, the msm6679al-110 does not sort si vocabularies for the best match, instead returning f63ah code when an utterance has been detected. when set to 0, normal recognition is performed. when this bit is set to 1, the host mcu can use the f343h command to upload the recognition parameter vector, so that the host can perform independent processing. xxxx xxx1 cleared f2xy = initialization acknowledge. [1] clear sd recognition and name tag ram. when set to 1, the msm6679al-110 initializes the sd parameter table. when set to 0, existing sd parameters are preserved. after this bit is set to 1, all sd training and name tag pointers are erased. use this command to start training for a new user. if the old name tags are to be retained, the f50ch command can recall old name tags from flash. to set up for a blank sd and name tag table at the next power-on, issue the command sequence f201h f507h.
msm6679al-110 voice recognition processor ? semiconductor 32 recognize opcode action return value msm6679al-110 was not in listen mode. none stop listening. this command causes the msm6679al-110 to exit si or sd listen mode, whichever was active. f301h - f33fh f300h aborting si listen mode. f600h aborting sd listen mode. f700h aborting si listen mode. f600h trigger detection code (see initialization command). f63ah rejection. f63bh~f63fh start si listen mode. for all the following opcodes, the msm6679al-110 per-forms si recognition on incoming utterances, using si vocabulary y. the vocabulary y is identified by one of 15 sets, thus y = 1h ~ fh. invalid signal processing table. f802h sample data overrun. f840h utterance id in vocabulary y. f6h utt return recognized phrase using vocabulary number y. utterance id in vocabulary y, high and low byte of distance to utterance 1... distance to utterance n. f6h utt dst1h dst1l... dstnh dstnl return recognized phrase and distance table for vocab y. utterance id in vocabulary y, high and low byte of minimum and maximum energy val-ue. f6h utt eminh eminl emaxh emaxl return recognized phrase and energy value for vocab y. f32yh f33yh f30yh f31yh utterance id, high and low byte of distance to utterance 1...distance to utterance n, high and low byte of minimum and maximum en-ergy value. f6h utt dst1h dst1l... dstnh dstnl eminh eminl emaxh emaxl return recognized phrase, distance table, and energy value for vocab y. triggered. f740 abort sd listen mode. f700 f340h rejection. f73e start sd listen mode. when an utterance is captured, it is analyzed and converted to a "recognition parameter vector." the host may then command the msm6679al-110 to use this vector in various ways (e.g., sort, update, or recognition vector upload). memory empty. f73f invalid sp table. f802 sample data overrun. f840 f341h, f344h, f351h, f361h, f371h abnormal response: memory empty. f73fh sd recognition sort. these commands sort the distances between the recognition parameter vector and the reference vectors for the utterances in the current sd vocabulary. utt = utterance id. f7h utt return recognized phrase for vocab y. this command can be issued several times to yield first, second, third best, etc. utt = index of recognized phrase, dsth dstl = high/low bytes of distance from nearest phrase. f7h utt dsth dstl return recognized phrase and distance for the current vocabulary. utterance id, high and low byte of distance to utt. 1...n. f7h utt dst1h dst1l... dstnh dstnl return recognized phrase and distance table for vocab y. utterance id, high and low byte of minimum and maximum energy value. f7h utt eminh eminl emaxh emaxl return recognized phrase and energy value for vocab y. f341h f344h f351 f361h
? semiconductor msm6679al-110 voice recognition processor 33 recognize (continued) speak opcode action return value f341h, f344h, f351h, f361h, f371h f371h utterance id, high and low byte of distance to utterance 1...distance to utterance n, high and low byte of minimum and maximum energy value. f7h utt dst1h dst1l... dstnh dstnl eminh eminl emaxh emaxl return recognized phrase, distance table, and energy value for vocab y. f342h update complete. f740h update sd recognition enrollment. this command updates enrollment on utter-ance utt, immediately after a "f7h utt" response to the sort sd distances command (f341h). alternatively, the utterance to be updated can be selected by the sd search command (f9xyh). this command uses the recognition parameter vector from the most recently captured utterance, and does not start sd listen mode. generally, update should be performed only if correct utterance identify is confirmed by the user. f343h f743h nh nl v1h v1l... vnh vnl = success, where nh/nl = high/low bytes of n, n = length of recognition parameter vector v, v1h/v1l = high/low bytes of first element of v, vnh/vnl = high/low bytes of nth element. recognition vector upload. request recognition parameter vector upload to host. failure. f743h 00 00 opcode f401h ~ f43dh action return value if enabled, this value is returned upon completion of playback. f400h speak phrase from external memory. this command causes the msm6679al-110 to play back a name tag from external memory. if no sound is defined for a selected index, the msm6679al-110 plays a beep. see the record commands for information on creating name tags. f441h ~ f450h if enabled, this value is returned upon completion of playback. f400h speak phrase from low internal memory. if no sound is defined for a selected index, the msm6679al-110 plays a beep. the default phrases supplied with the msm6679al-110 in the smaller low playback memory area are listed below. buzzer. dial tone. bonk. f442h drip. f441h f443h f444h
msm6679al-110 voice recognition processor ? semiconductor 34 speak (continued) opcode action return value f451h ~ f47ch if enabled, this value is returned upon completion of playback. f400h speak phrase from high internal/external memory. if no sound is defined for a selected index, the msm6679al-110 plays a beep. the default phras-es supplied with the msm6679al-110 in the larger upper playback memory area are listed below. "0" simulated dtmf tone. "1" simulated dtmf tone. "2" simulated dtmf tone. "3" simulated dtmf tone. "4" simulated dtmf tone. "5" simulated dtmf tone. "6" simulated dtmf tone. "7" simulated dtmf tone. "8" simulated dtmf tone. "9" simulated dtmf tone. "*" simulated dtmf tone. f452h f451h f453h f454h f455h f456h f457h f458h f459h f45ah f45bh f45ch "#" simulated dtmf tone. f47d reserved. this command is reserved for future use. f47eh if enabled, this value is returned upon completion of playback. f400h beep. this causes the msm6679al-110 to beep for 50 ms. f47fh if enabled, this value is returned upon completion of playback. f400h pause. this command can be issued while the msm6679al-110 is performing sound output and is then put in the msm6679al-110 command stack for subsequent processing. when this command is executed, sound output pauses for 0.2 sec. the pause command is useful for word spacing. f480h none. set msm6654 mode. this command causes the msm6679al-110 to initialize the external msm665x device, also clearing the device from busy mode. f481h - f4ffh if enabled, this value is returned upon completion of playback. if nar is set, the f400h command is sent when the msm665x device is ready for an-other command. if busy mode is selected, the f400 command is returened when the sound is finished. f400h playback sound from msm665x device. this command causes the msm6679al-110 to issue a speak command to the msm665x slave device. the value is passed on the msm665x device as 01h - 07fh. the actual phrase is determined by the vocabulary programmed into the msm665x device. up to 127 external phrases are supported. f50bh none. set msm665x busy mode on.
? semiconductor msm6679al-110 voice recognition processor 35 speak (continued) request opcode action return value f51bh none. set 6654 nar mode. this command, which is the complement of the f50b command, sets up the handshaking to the attached 6654 speech synthe-sizer to use the nar. this setup uses the 6654's double buffer feature to eliminate any gap between two consecutive phases. fexyh none. set output level. this command sets the speech output level to one of 255 values as follows: fe03 set minimum output level. fe80h set output level half way (default). fefeh set maximum output level. opcode action return value msm6679al-110 ready. f500h msm6679al-110 disabled. f520h msm6679al-110 waiting for start. f540h f500h msm6679al-110 waiting for end. f560h status request. this command causes the msm6679al-110 to return a 2-byte value indicating its current status. msm6679al-110 processing. f580h download/upload in progress. f5a0h download/upload complete. f5c0h select/jump complete. f5e0h f501h select last flash bank for si recognition. f510h no return value select download ram bank for si/sp template area. this command enables the download ram bank in the upper 32 k of data memory for si recognition. f520h no return value select buffer ram bank for si/sp. this command enables the buffer ram bank in the upper 32 k of data memory for si recognition.
msm6679al-110 voice recognition processor ? semiconductor 36 request (continued) opcode action return value f502h immediately after receiving parameter nl, the msm6679al-110 responds with a message to indicate acceptance or denial of the transfer request. acceptance is indicated by f5a0h. denial is indicated by a f8xyh. at the end of an accepted transfer, the msm6679al-110 re-sponds with a message to confirm or deny valid completion of the transfer. valid completion is indicated by f5c0h. download/upload. full syntax: f5 02 00 ctl adh adl nh nl [dt1... dtn [dt(n+1)]] full syntax: f5 02 00 ctl adh adl nh nl [dt1... dtn [dt(n+1)]] ctl(7) = 0 for download, ctl(7) = 1 for upload ctl(6) = 0 for data ram, ctl(6) = 1 for program ram/rom if ctl(6)=0 then ctl(1-0) = seg: data segment selection if ctl(6)=1 and ctl(1-0) = x0, then external program segment 0 is used. if ctl(6)=1 and ctl(1-0) = x1, then external program segment 1 is used. adh adl = high, low bytes of starting address. nh nl = high, low bytes of n n = number of bytes to be downloaded or uploaded (maximum 07ffch) dt1... dtn = download data. note (here and in upload response) that data are 8-bit binary values, even if using the serial interface. dt(n+1). if n is odd, an extra byte is appended to the data so that the total number of bytes in the message remains even. this command requests data transfer to/from data or external program memory.the control parameter (ctl) controls the direction of the transfer (i.e., download vs. upload) and specifies which of six 64-kbyte memory segments (i.e., four data segments and two external program segments) is to be accessed. this command does not work with internal program memory. it is not possible to download to external program memory while running in external program memory. the address and length parameters (adh adl nh nl) specify the starting address and length of the transfer in bytes. since the msm6679al-110 can only perform download /upload transfers within one 32-kbyte block in one download /upload command, the address and length parameters must not specify a transfer that violates a 32-kbyte address boundary. if this restriction is violated, the download/upload request will be denied. invalid message received. f880h sample data over-run. f840h 32-kbyte block boundary violation error. f820h unclassified download/upload error. f810h divide-by-zero error. f808h select/jump error. f804h invalid sp header or table. f802h reserved. f801h most and least significant byte of ad-dress where error occurred. faxyh fbxyh f504h four-digit ascii number. xxxx retrieve msm6679al-110 firmware revision number.
? semiconductor msm6679al-110 voice recognition processor 37 request (continued) opcode action return value f505h initialization is complete. f501h initialize in background. background noise initialization is performed for 500 ms. the msm6679al-110 requires this command prior to recognition for noise vector subtraction during the utterance sampling period. use the background initialization command whenever there is a change in the background noise level. for example, sample the noise signature in a vehicle at rest and moving at 35 mph with its windows rolled down. the quality of a phone line connection can also vary from call to call. the host mcu must implement a strategy as to when to issue a background initialization command. in a vehicle, the host mcu could monitor the vehicle speed, fan speed, radio volume, etc. alternatively, the host mcu could issue this command each time a new recognition session starts or a new line connection is established. however, the 0.5-sec sample period could degrade system responsiveness if used too frequently. a zero in this bit location during the f2xxh command will not cause an initialization. the f2xxh command can also be used to perform background noise initialization. f506h four digit ascii number. xxxx retrieve vocabulary and trigger table revision number. f507h save is complete. f501h save sdr templates in last flash. save the download ram bank sd template area. saves 2684 bytes from the address set by the f103 command to the address range f300- fd7f in the last flash. the default is 4a00- 547b ? f300-fd7f). f508h no return value get sdr templates from last flash. get the download ram bank sd template area. saves 2684 bytes to the address set by the f103 command from the address range f300- fd7b in the last flash. the default is (f300- fd7b ? 4a00-547b). f509h select default si vocabulary. (first flash)
msm6679al-110 voice recognition processor ? semiconductor 38 record opcode f101h 00xxh action return value operation complete. f101h 00xxh set name tag length, set msm665x busy mode on. name tag record length is set by xxh, with xxh defining record length in 14-ms intervals. the maximum record length of ffh yields a recording interval of 3.57 sec. the default value is 1.2 sec. f105 xxxx f105 baaa, where b is the bank num-ber (0,1,2), and aaa is the bank ad-dress /16 (800 - ff8) set name tag record origin. this command sets the beginning address for recording name tags. xxxx = 128 byte blocks from 0000 to 02ff. the reset default is 0000. this is only effective before an f50a command since new recordings start after the end of the previous recording. the f50a command uses this num-ber to calculate the first address. f106 xxxx f106 baaa, where b is the bank num-ber (0,1,2), and aaa is the bank ad-dress /16 (800 - ff8) set name tag record end. this command sets the ending address for recording name tags. xxxx = 128 byte blocks from 0000 to 02ff. the reset default is 01ff. f50ah name tag table cleared. f501h clear name tag table. f50ch saved name tag table recalled. f501h recall name tag pointers from first flash. save the first flash name tag pointers (fd80 - ffff) to the working name tag pointer table. the default is (fd80-ffff ? 5480-56ff). f51ch name tag pointers recalled. f501h recall name tag pointers from last flash. save the last flash name tag pointesr (fd80 - ffff) to the working name tag pointer table. the default is (fd80-ffff ? 5480-56ff). f50dh name tag table saved. f501h save name tag pointers in first flash. save the working name tag pointer table to the first flash name tag pointers. the default is (5480 -56fd ? fd80-fffd). f51dh name tag pointers saved. f501h save name tag pointers in last flash. save the working name tag pointer table to the last flash name tag pointers. the default is (5480 -56fd ? fd80-fffd). f50eh set record volume high. f50fh set record volume to normal. this is the default setting. fa00h reserved. this command is reserved for future use. fa01h ~ fa3dh completed. fa00h record name tag. memory full. f280h
? semiconductor msm6679al-110 voice recognition processor 39 record (continued) sd recognition control opcode action return value fa3dh ~ faffh reserved. these commands are reserved for future use. opcode action return value recognition performance is largely a function of how well the enrollment data represents subsequent tokens of the enrolled utterances, and performance generally improves steadily with each additional enrollment pass. for most applications, three initial enrollment passes are recommended. subsequent reference updating can be performed with the sd recognize update command (f342). f521h sdr table is cleared f501h clear sdr table. this command initializes a blank sd template table. the 2684-byte area from the address set by the f103 command (the working sdr table) is set to zeros. the sdr tables in the flash banks are not affected. the default is (4a00 - 547b). f6xyh no return value. set sd segment pointer. this command sets the sd segment pointer to xy00h, i.e., set the starting address of the current sd recognition parame-ter table to xy00h. issuing this command is equivalent to issuing the set sd origin command, f103h xy00h. (for further details of operation, please refer to the description of that command.) utterance number found. f740h f9xyh utterance number not found. f700h search for sd utterance xy. this is the first step in adding an utterance to the vocabulary, or in replacing an existing one. the sd vocabulary memory is searched for utt. no. xyh. if it is not found and if sufficient sd memory exists, the msm6679al-110 prepares to add utterance number xyh to the vo-cabulary. memory full. f73fh operation complete. f740h aborting sd listen mode. f700h fb00h improper level, must repeat. f73eh enroll sd utterance. this command starts msm6679al-110 sd listen mode, then uses the next captured utterance to start or update training of the reference data for sd utterance number xy specified in the most recent search command (f9xyh). the user must be prompted to say the utter-ance prior to issuing this command. if the utterance was previously enrolled, a training update is performed; if not, the reference data is initialized. each utterance in the sd vocabulary must be enrolled at least once before it can be recognized. invalid signal processing table. f802h sample data overrun. f840h fc00h operation complete. f740h erase utterance from sd vocabulary. this command erases the reference parameters for utterance number xyh from the sd vocabulary, where xyh is the utterance number retained from the previous search command (f9xyh).
msm6679al-110 voice recognition processor ? semiconductor 40 asynchronous serial protocol example all messages to the msm6679al-110 (except downloads and uploads) are echoed, but replies from the msm6679al-110 to the host are not echoed by the host. this arrangement facilitates manual communication with the msm6679al-110 using standard terminals. the following table illustrates the range of msm6679al-110 functions. comment action initialize msm6679al-110 host initializes msm6679al-110. msm6679al-110 acknowledges. msm6679al-110 response host command voice input f258 f200 f258 load trigger tables at 5000h. host requests download to data segment 0, starting at location 5000h, of 256 bytes (0100h). msm6679al-110 accepts request. host sends 256 bytes (~0.25 sec at 9600 baud). msm6679al-110 indicates download complete. f502 0000 5000 0100 f5a0 f5c0 f502 0000 5000 0100 ... set new triggering origin. host requests set triggering origin to 5000h. msm6679al-110 sets triggering origin and sends confirming response. f104 5000 f104 5000 f104 5000 download new sd vocabulary. host requests download to data segment 0, starting at location 6000h, of 4 kbytes (1000h). msm6679al-110 accepts request. host sends 4 kbytes (~4.3 sec at 9600 baud) msm6679al-110 indicates download complete. f502 0000 6000 1000 f5a0 f5c0 f502 0000 6000 1000 ...
? semiconductor msm6679al-110 voice recognition processor 41 comment action set new sd tables. host requests set sd origin to 6000h. msm6679al-110 sets sd origin and responds. msm6679al-110 response host command voice input f103 6000 f103 6000 f103 6000 download first 4 k of si vocabulary. host requests download to data segment 0, starting at location 7000h, of 4k bytes (1000h). msm6679al-110 accepts request. host sends 4 kbytes. msm6679al-110 indicates download complete. f502 0000 7000 1000 f5a0 f5c0 f502 0000 7000 1000 ... download last 32 k of si vocabulary. host requests download to data segment 0, starting at location 8000h, of 32k bytes (7ffc). msm6679al-110 accepts request host sends 32 kbytes. msm6679al-110 indicates download complete. f502 0000 8000 7ffc f5a0 f5c0 f502 0000 8000 7ffc ... set new sp/si tables. host requests set sp/si origin = 7000h. msm6679al-110 sets sp/si origin and responds. f102 7000 f102 7000 f102 7000 upload data for diagnostics. host requests upload from data segment 0, starting at location 300h, of 45 bytes (2dh). msm6679al-110 accepts request, signals in progress. msm6679al-110 sends 46 bytes. msm6679al-110 indicates upload complete. f502 00a0 0300 002d f5a0 ... f5c0 f502 00a0 0300 002d set up msm6679al-110 for si recognition. host requests set sp table 3. msm6679al-110 selects sp table 3 and confirms. host initializes msm6679al-110. msm6679al-110 acknowledges. f123 f123 f258 f200 f123 f258 si recognition. host starts si recognition, vocabulary 1. msm6679al-110 recognizes utterance 3. host starts si recognition, vocabulary 2. msm6679al-110 recognizes utterance 2. host starts si recognition, vocabulary 2. msm6679al-110 recognizes utterance 3. f301 f603 f302 f602 f302 f603 f301 f302 f302 "dial" "two" "three"
msm6679al-110 voice recognition processor ? semiconductor 42 comment action si recognition. host starts si recognition, vocabulary 1. msm6679al-110 recognizes utterance 1. msm6679al-110 response host command voice input f301 f601 f301 "store" si recognition of control words. host starts si recognition, vocabulary 1. msm6679al-110 recognizes utterance 3. host starts si recognition, vocabulary 2. msm6679al-110 recognizes utterance 5. host starts si recognition, vocabulary 2. msm6679al-110 recognizes utterance 6. host starts si recognition, vocabulary 1. msm6679al-110 recognizes utterance 7. f301 f603 f302 f605 f302 f606 f301 f601 f301 f302 f302 f301 "dial" "five" "six" "store" sd enrollment. host prepares msm6679al-110 to train sd utterance 2 memory is empty and ready to train. pass 1; host sends sd enroll command. sd utterance 2 initialized. pass 2; host sends sd enroll command. msm6679al-110 updates sd utterance 2. pass 3; host sends sd enroll command. msm6679al-110 signals operation completed. f902 f700 fb00 f740 fb00 f740 fb00 f740 f902 fb00 fb00 fb00 "bill jones" "bill jones" "bill jones" si recognition of control word. host starts si recognition, vocabulary 1. msm6679al-110 recognizes utterance 3. f301 f603 f301 "directry" sd recognition. host starts sd recognition. msm6679al-110 signals trigger ok. host sends sd sort command. msm6679al-110 recognizes utterance 1. f340 f740 f341 f701 f340 f341 "john smith" sd enrollment. get ready to train sd utterance 1. memory is empty and ready to train. pass 1; host sends sd enroll command. sd utterance 1 initialized. pass 2; host sends sd enroll command. sd utterance 1 updated. pass 3. host sends sd enroll command. sd utterance 1 updated. "john smith" "john smith" "john smith" f901 fb00 fb00 fb00 f901 f700 fb00 f740 fb00 f740 fb00 f740
? semiconductor msm6679al-110 voice recognition processor 43 comment action name tag recording. host initiates msm665x port. host sets recording length to 1 sec. msm6679al-110 signals operation complete. host clears name tag table msm6679al-110 signals operation complete. host sets record gain to max. level. start recording tag one. msm6679al-110 signals name tag recording complete. save name tags to flash. name tags saved. msm6679al-110 response host command voice input f480 f101 0047 f101 0047 f50a f501 f50e fa01 fa00 f50d f501 f480 f101 0047 f50a f50e fa01 f50d "jane doe" name tag playback. host sets volume to max. level. host commands play back name tag 1. msm6679al-110 signals playback ok. feff f401 "jane doe" f400 feff f401 sound playback. host sets output volume to mid point. play msm6679al-110 internal sound 1. play back sound from msm6654. fe80 f442 "bzzzz" f49f "completed" fe80 f442 f49f
msm6679al-110 voice recognition processor ? semiconductor 44 notice 1. the information contained herein can change without notice owing to product and/or technical improvements. before using the product, please make sure that the information being referred to is up-to-date. 2. the outline of action and examples for application circuits described herein have been chosen as an explanation for the standard action and performance of the product. when planning to use the product, please ensure that the external conditions are reflected in the actual circuit, assembly, and program designs. 3. when designing your product, please use our product below the specified maximum ratings and within the specified operating ranges including, but not limited to, operating voltage, power dissipation, and operating temperature. 4. oki assumes no responsibility or liability whatsoever for any failure or unusual or unexpected operation resulting from misuse, neglect, improper installation, repair, alteration or accident, improper handling, or unusual physical or electrical stress including, but not limited to, exposure to parameters beyond the specified maximum ratings or operation outside the specified operating range. 5. neither indemnity against nor license of a third party's industrial and intellectual property right, etc. is granted by us in connection with the use of the product and/or the information and drawings contained herein. no responsibility is assumed by us for any infringement of a third party's right which may result from the use thereof. 6. the products listed in this document are intended for use in general electronics equipment for commercial applications (e.g., office automation, communication equipment, measurement equipment, consumer electronics, etc.). these products are not authorized for use in any system or application that requires special or enhanced quality and reliability characteristics nor in any system or application where the failure of such system or application may result in the loss or damage of property, or death or injury to humans. such applications include, but are not limited to, traffic and automotive equipment, safety devices, aerospace equipment, nuclear power control, medical equipment, and life-support systems. 7. certain products in this document may need government approval before they can be exported to particular countries. the purchaser assumes the responsibility of determining the legality of export of these products and will take appropriate and necessary steps at their own expense for these. 8. no part of the contents cotained herein may be reprinted or reproduced without our prior permission. printed in japan e2y0001-28-41 copyright 1998 oki electric industry co., ltd.


▲Up To Search▲   

 
Price & Availability of MSM6650

All Rights Reserved © IC-ON-LINE 2003 - 2022  
[Add Bookmark] [Contact Us] [Link exchange] [Privacy policy]
Mirror Sites :  [www.datasheet.hk]   [www.maxim4u.com]  [www.ic-on-line.cn] [www.ic-on-line.com] [www.ic-on-line.net] [www.alldatasheet.com.cn] [www.gdcy.com]  [www.gdcy.net]
 . . . . .
  We use cookies to deliver the best possible web experience and assist with our advertising efforts. By continuing to use this site, you consent to the use of cookies. For more information on cookies, please take a look at our Privacy Policy. X