Post by thread:Generating Digital Speech cheaply

I would like to hear some opinions from other people that have already done the same thing I plan to do.... Basically I want to make a very cheap voice playback device capable of playing back a fixed set number of short pre-recorded voice messages. I know you get special IC's and CODECS that does exactly this, but they are still slightly too expensive for what I had in mind. EPROM IC's are very inexpensive these days, and I thought that maybe if store my messages as raw digitized speech on a large EPROM and play it back through a resistor ladder network A/D and filter, I would be able to regenerate speech messages at a relatively low cost. If I then use a PIC to address the EPROM and add a serial interface to it, this could become a very cheap speech messaging system that is addressable via the serial bus from my target microcontroller. Now, my questions are this: 1. Is it a feasible way of doing this? 2. What sampling rate would I need? (I was thinking 8KHz @ 8 bits resolution) Is this good enough? 3. Has anybody perhaps done something like this already? Was the sound reasonably clear? One final question.... Some time ago, someone on the Piclist mentioned a stand-alone text-to-speech synthesizer available for as low a cost as $49.95. I can not remember the details anymore. Do anybody where I can buy one of these? I made a search on the web, but all I could come up with is a device from RC systems http://members.aol.com/rcsys which costs about $150, and is thus a bit too expensive for what I had in mind. Have anybody reading this ever used these text-to-speech modules and do they work well? Rgds Werner -- Werner Terreblanche <spam_OUTwterrebTakeThisOuTplessey.co.za>Tel : (021) 7102251 (office hours)

Werner, I would look at an OKI chip (MSM5202) that does 4 bit ADPCM encoded speech. at 8 khz it sounds really good. and don't forget that's only 4 bits so each byte is 2 samples. record your message with a PC into a WAV file and OKI has software that will translate that to 4 bit ADPCM. Stuff that file in EPROM and it would be easy for the PIC to feed it to the OKI chip at 8khz. the OKI chip provides an interrupt to ask for another sample. I did a similar design many years ago.... good luck. Jason Wolfson Lipidex Corp {Original Message removed}

> Basically I want to make a very cheap voice playback device capable > 1. Is it a feasible way of doing this? > 2. What sampling rate would I need? (I was thinking 8KHz @ 8 bits > resolution) Is this good enough? > 3. Has anybody perhaps done something like this already? Was the sound > reasonably clear? I did this with an 8051 (not that it makes any difference) a while ago and it worked well. One of the Windows WAV file utilities lets you play with bit sizes and sample rates, so you can try it on your PC and see if it sounds good enough. It will show you the data size too, if you are planning on saving the data as a WAV file in your ROM which is about as easy as it gets. Steve. ====================================================== Steve Baldwin Electronic Product Design TLA Microsystems Ltd Microcontroller Specialists PO Box 15-680, New Lynn http://www.tla.co.nz Auckland, New Zealand ph +64 9 820-2221 email: .....stevebKILLspam@spam@tla.co.nz fax +64 9 820-1929 ======================================================

Werner Terreblanche wrote: {Quote hidden}> > I would like to hear some opinions from other people that have already done > the same thing I plan to do.... > > Basically I want to make a very cheap voice playback device capable of > playing back a fixed set number of short pre-recorded voice messages. I > know you get special IC's and CODECS that does exactly this, but they are > still slightly too expensive for what I had in mind. EPROM IC's are very > inexpensive these days, and I thought that maybe if store my messages as raw > digitized speech on a large EPROM and play it back through a resistor ladder > network A/D and filter, I would be able to regenerate speech messages at a > relatively low cost. If I then use a PIC to address the EPROM and add a > serial interface to it, this could become a very cheap speech messaging > system that is addressable via the serial bus from my target > microcontroller. > > Now, my questions are this: > > 1. Is it a feasible way of doing this? > 2. What sampling rate would I need? (I was thinking 8KHz @ 8 bits > resolution) Is this good enough? > 3. Has anybody perhaps done something like this already? Was the sound > reasonably clear? > > One final question.... Some time ago, someone on the Piclist mentioned a > stand-alone text-to-speech synthesizer available for as low a cost as > $49.95. I can not remember the details anymore. Do anybody where I can > buy one of these? I made a search on the web, but all I could come up with > is a device from RC systems http://members.aol.com/rcsys which costs about > $150, and is thus a bit too expensive for what I had in mind. Have anybody > reading this ever used these text-to-speech modules and do they work well? > > Rgds > Werner > -- > Werner Terreblanche <wterrebKILLspamplessey.co.za>Tel : (021) 7102251 (office > hours) Werner: hard to beat the I.S.D.(information storage devices) chips ie- [isd 2560]= 60 second message recorder; addressable/standalone/multi cue-play features approx $20.00 (NZ) has mic & 16ohm spkr inp/drv on chip already, 28 pin. uses multi level programing of internal eeprom for hi-density. should save a lot of real estate. regards Graham Daniel.

You have a lot of choice to do this project (isd recored/playback chip or oki speech IC etc etc ) One thing I am not sure about is the PIC...I have tried to interface Pic with ISD chipcorder for my talking clock project but with out success...if you are familiar with motorola uC I would say go for motorola. by the way, if all you need is a short message to play back ISD chipcorder is the best choice around. S+ {Quote hidden}> Werner, > I would look at an OKI chip (MSM5202) > that does 4 bit ADPCM encoded speech. at 8 khz it sounds really good. and > don't forget that's only 4 bits so each byte is 2 samples. > record your message with a PC into a WAV file > and OKI has software that will translate that to > 4 bit ADPCM. Stuff that file in EPROM and it would > be easy for the PIC to feed it to the OKI chip at 8khz. > the OKI chip provides an interrupt to ask for another sample. > I did a similar design many years ago.... > good luck. > Jason Wolfson > Lipidex Corp > > {Original Message removed}

On Mon, 20 Apr 1998 12:35:21 +0200 Werner Terreblanche <.....wterrebKILLspam.....PLESSEY.COM> writes: >I would like to hear some opinions from other people that have already >done >the same thing I plan to do.... > >Basically I want to make a very cheap voice playback device capable of >playing back a fixed set number of short pre-recorded voice messages. >I >know you get special IC's and CODECS that does exactly this, but they >are >still slightly too expensive for what I had in mind. EPROM IC's are >very >inexpensive these days, and I thought that maybe if store my messages >as raw >digitized speech on a large EPROM With any storage method, the #1 thing is to get a good original recording of the speech to begin with. Some geek speaking into a microphone built into the front panel of a noisy computer just isn't going to sound good, no matter what storage method is used. If the total amount of speech is 60 seconds or less the analog-store chips such as ISD's are not a bad choice. They have enough internal control logic that it may be possible to get away without the PIC. In any case, they won't need intervention from the PIC once they start playing. Stay away from the ones that sample at less than 6.4 KHz. If you take "raw samples" they should be u-law or A-law for better dynamic range. Or compress the speech before recording, then a linear converter will work for playback of course it's still compressed. It can be hard to make a "resistor tree" DAC work for more than 6 bits or so. I'm still looking for where to buy the hybrid ones that are occasionally found in consumer products. Note that raw 8-bit samples at 8 KHz need 64 Kbit/sec to store, or 3.84 Mbit/minute. A 4 MBit EPROM will hold only about 64 seconds of such speech. I had good results using CVSD techniques. Only one bit per sample is taken, but the sample rate needs to be about 32 KHz before it starts to sound good (this is half the bit rate of raw PCM samples). Rates of 50 KHz or more sound really good. The decoder can be built with a few lines of PIC software, a few op-amps and an analog switch. The decoder circuit can also be re-used as the core of an encoder circuit. No special ICs are required. I'm sure there's an easy way to convert PCM to CVSD on a PC but I haven't tried yet. _____________________________________________________________________ You don't need to buy Internet access to use free Internet e-mail. Get completely free e-mail from Juno at http://www.juno.com Or call Juno at (800) 654-JUNO [654-5866]

Steve Thanks for your reply. Inadvertantly, you also answered another question that I had... namely "Is it possible to store WAV files directly in ROM and play it back?" I have never been able to confirm how a WAV file is encoded. Are you saying it is possibly to store the WAV file directly in ROM and play it back just like that through a D->A converter? Rgds Werner -- Werner Terreblanche <EraseMEwterrebspam_OUTTakeThisOuTplessey.co.za>Tel : (021) 7102251 (office hours) {Quote hidden}> ---------- > From: Steve Baldwin[SMTP:stevebspam_OUTKCBBS.GEN.NZ] > Sent: Tuesday, April 21, 1998 2:44 AM > Subject: Re: Generating Digital Speech cheaply > > > Basically I want to make a very cheap voice playback device capable > > 1. Is it a feasible way of doing this? > > 2. What sampling rate would I need? (I was thinking 8KHz @ 8 bits > > resolution) Is this good enough? > > 3. Has anybody perhaps done something like this already? Was the sound > > reasonably clear? > > I did this with an 8051 (not that it makes any difference) a while > ago and it worked well. One of the Windows WAV file utilities lets > you play with bit sizes and sample rates, so you can try it on your > PC and see if it sounds good enough. It will show you the data size > too, if you are planning on saving the data as a WAV file in your > ROM which is about as easy as it gets. > > Steve. > > > ====================================================== > Steve Baldwin Electronic Product Design > TLA Microsystems Ltd Microcontroller Specialists > PO Box 15-680, New Lynn http://www.tla.co.nz > Auckland, New Zealand ph +64 9 820-2221 > email: @spam@stevebKILLspamtla.co.nz fax +64 9 820-1929 > ====================================================== >

Thank you for your advice Mike Keitz, Sam Haile, Graham Daniel, Steve Baldwin and Jason Wolfson. I suppose it makes a lot of sense to investigate the ISD chips a little bit more closely. The things that I need to find out about these chips are if their memories are non-volatile and whether it is easy to make duplicates of pre-recorded chips for production purpouses. The ones I've seen so far does not meet either criteria. I once saw a Texas Instruments chip once which meets both these criteria. It's a one-time programmable chip which allows you to store a number of speech messages (128 seconds worth). The disadvantage is that its one time programmable, but the advantage is that it's completely non-volatile. Decisions... decisions hey? :) Rgds Werner

On Tue, 21 Apr 1998 08:22:03 +0200 Werner Terreblanche <TakeThisOuTwterrebEraseMEspam_OUTPLESSEY.COM> writes: >Steve > > Thanks for your reply. Inadvertantly, you also answered another >question >that I had... namely "Is it possible to store WAV files directly in >ROM and >play it back?" I have never been able to confirm how a WAV file is >encoded. It's fairly widely documented on the net. The format is rather flexible. Each "chunk" of data has a header indicating the encoding method, number of bits, number of channels, sample rate, etc. Then the rest of the "chunk" is just the data packed in into as many bytes per sample as needed. If the encoding is just mono 8-bit PCM, then each sample is one byte, so you could just remove the header and store the rest directly. In most cases, the entire file is one "chunk" so the only header is at the beginning. There is also provision for non-audio "chunks" like the title, author, copyright, note, etc. You'd need to remove those, or be sure not to store them in the first place. Are you saying it is possibly to store the WAV file >directly in >ROM and play it back just like that through a D->A converter? The 8-bit formats are nonlinear PCM (u-law) which probably won't sound right played back through a linear converter. Other than that it would work. _____________________________________________________________________ You don't need to buy Internet access to use free Internet e-mail. Get completely free e-mail from Juno at http://www.juno.com Or call Juno at (800) 654-JUNO [654-5866]

On Tue, 21 Apr 1998 08:58:44 +0200 Werner Terreblanche <RemoveMEwterrebTakeThisOuTPLESSEY.COM> writes: >I suppose it makes a lot of sense to >investigate >the ISD chips a little bit more closely. The things that I need to >find out >about these chips are if their memories are non-volatile Yes it is, they are EEPROM. They claim 40 to 100 years data retention and it seems that the chips (at least the ones with the parallel interface) are well-protected against unintentional writing by tying the R/W pin high. and whether >it is >easy to make duplicates of pre-recorded chips for production >purpouses. It's not hard, but the normal recording mode must be used (meaning it will take 20 seconds to duplicate a 20 second chip, etc.). ISD offers a "programmer" that makes duplicates or one could be easily built. If you need tens of thousands of duplicates the duplication may be a problem. If that is your case, write or call ISD and ask them if they can supply pre-programmed chips. They may be willing to work with you if you are dealing in large quantities. >I once saw a Texas Instruments chip once which meets both these >criteria. >It's a one-time programmable chip which allows you to store a number >of >speech messages (128 seconds worth). The disadvantage is that its one >time >programmable, but the advantage is that it's completely non-volatile. >Decisions... decisions hey? :) I think the TI chips use a LPC algorithm to store a lot of speech in a small memory, but with a very heavy "synthesizer" effect on the sound. If this is the case, you'll need special software which TI may not give you in order to encode your speech for programming. But a large library of common standard words should be available. I was just looking at ISSI (http://www.issiusa.com). They have some chips which store ADPCM digitally in a OTP ROM. The memory capacity isn't very large (512K bit, samples are 4 bits so that's 16 seconds at 8K sampling) but I think some of them had interfaces for external ROM. I don't know if the quality would be very good either since they seem to be targeted to very cheap "squawking" devices using piezo speakers. Everything about the ISSI chips (how to encode, how to program) seems to be proprietary. _____________________________________________________________________ You don't need to buy Internet access to use free Internet e-mail. Get completely free e-mail from Juno at http://www.juno.com Or call Juno at (800) 654-JUNO [654-5866]

yes http://www.isd.com On Tue, 21 Apr 1998, Paul Dartanian wrote: {Quote hidden}> Do you know a website for this "ISD chipcorder analog EPROM > > > -----Original Message----- > From: Haile, Sam <shaileEraseME.....ESSEX.AC.UK> > To: EraseMEPICLISTMITVMA.MIT.EDU <RemoveMEPICLISTEraseMEEraseMEMITVMA.MIT.EDU> > Date: Tuesday, April 21, 1998 12:07 PM > Subject: Re: Generating Digital Speech cheaply > > > >ISD chipcorder is an analog EPROM "NOT DIGITAL" you can record > >message again and again. > > > > it is nonvolatile. > > > >sam > > >

On Tue, 21 Apr 1998 17:45:51 +0300 Paul Dartanian <RemoveMEpauligspam_OUTKILLspamcyberia.net.lb> writes: >Do you know a website for this "ISD chipcorder analog EPROM http://www.isd.com The complete name of the company is "Information Storage Devices". Their name for the process used in the chipcorders is "DAST" (Direct Analog Storage Technology). Searching on these terms as well as the obvious ones like "ISD", "Chipcorder", etc. may turn up some relatee third-party web sites. But the ISD site has a lot of information including data pdf's, theories of operation, and even I think some application notes. _____________________________________________________________________ You don't need to buy Internet access to use free Internet e-mail. Get completely free e-mail from Juno at http://www.juno.com Or call Juno at (800) 654-JUNO [654-5866]

> Thanks for your reply. Inadvertantly, you also answered another question > that I had... namely "Is it possible to store WAV files directly in ROM and > play it back?" I have never been able to confirm how a WAV file is > encoded. Are you saying it is possibly to store the WAV file directly in > ROM and play it back just like that through a D->A converter? Pretty much so, yes. There are a few variations of the standard WAV file and the type is encoded in a header section. In my case, I was able to stipulate that the sound would be 8bit, 11kHz so I didn't need the header. When I got the file, I loaded it into the Windows thing, saved it with those settings and then chopped off the header in a binary editor. The header even tells you how much to chop off. The header is pretty straight forward (sample rate, mono/stereo, 8/16 bit, file length, etc) and the data is the straight binary value. Read it out of ROM, write it to DAC. Someone else has already posted a link for a file format page. There is a very long document (on ibm's web site I think) that has the full RIFF specification. Don't bother with it unless you want to use APCM (which is a lot more work but less ROM). BTW. The ISD devices store raw digitised values. Some speech recording chips compress the sound in a way that really only works with speech. If you are storing sound effects, they sound awful. The ISD ones are OK for that. Steve. ====================================================== Steve Baldwin Electronic Product Design TLA Microsystems Ltd Microcontroller Specialists PO Box 15-680, New Lynn http://www.tla.co.nz Auckland, New Zealand ph +64 9 820-2221 email: RemoveMEstevebTakeThisOuTspamtla.co.nz fax +64 9 820-1929 ======================================================

> The 8-bit formats are nonlinear PCM (u-law) which probably won't sound > right played back through a linear converter. Other than that it would > work. No they aren't. Well, I should say that's not entirely correct. The simplest 8 bit format is straight 8 bit data. There is the option to save as u-Law, A-Law, ADPCM, and a host of other formats, but what Microsoft call PCM is straight out of the ADC. Calling it PCM is a bit of a misnomer anyway. While it's in the file, it's 8 bit (parallel) data. >From memory, the reason I didn't use the ISD device is that with the larger parts you couldn't concatenate messages without a small pause between them. I needed a long, continuous recording. Steve. ====================================================== Steve Baldwin Electronic Product Design TLA Microsystems Ltd Microcontroller Specialists PO Box 15-680, New Lynn http://www.tla.co.nz Auckland, New Zealand ph +64 9 820-2221 email: EraseMEstevebspamspamBeGonetla.co.nz fax +64 9 820-1929 ======================================================

Graham (and Werner), ISD recently announced a 16 min version of their 4000 series chips. I'm not sure of the price. - Tom At 12:49 AM 4/21/98 +1200, Graham Daniel wrote: {Quote hidden}>Werner Terreblanche wrote: >> >> I would like to hear some opinions from other people that have already done >> the same thing I plan to do.... >> >> Basically I want to make a very cheap voice playback device capable of >> playing back a fixed set number of short pre-recorded voice messages. I >> know you get special IC's and CODECS that does exactly this, but they are >> still slightly too expensive for what I had in mind. EPROM IC's are very >> inexpensive these days, and I thought that maybe if store my messages as raw >> digitized speech on a large EPROM and play it back through a resistor ladder >> network A/D and filter, I would be able to regenerate speech messages at a >> relatively low cost. If I then use a PIC to address the EPROM and add a >> serial interface to it, this could become a very cheap speech messaging >> system that is addressable via the serial bus from my target >> microcontroller. >> >> Now, my questions are this: >> >> 1. Is it a feasible way of doing this? >> 2. What sampling rate would I need? (I was thinking 8KHz @ 8 bits >> resolution) Is this good enough? >> 3. Has anybody perhaps done something like this already? Was the sound >> reasonably clear? >> >> One final question.... Some time ago, someone on the Piclist mentioned a >> stand-alone text-to-speech synthesizer available for as low a cost as >> $49.95. I can not remember the details anymore. Do anybody where I can >> buy one of these? I made a search on the web, but all I could come up with >> is a device from RC systems http://members.aol.com/rcsys which costs about >> $150, and is thus a bit too expensive for what I had in mind. Have anybody >> reading this ever used these text-to-speech modules and do they work well? >> >> Rgds >> Werner >> -- >> Werner Terreblanche <RemoveMEwterrebKILLspamplessey.co.za>Tel : (021) 7102251 (office >> hours) >Werner: hard to beat the I.S.D.(information storage devices) chips ie- >[isd 2560]= 60 second message recorder; addressable/standalone/multi >cue-play features approx $20.00 (NZ) has mic & 16ohm spkr inp/drv on >chip already, 28 pin. uses multi level programing of internal eeprom for >hi-density. should save a lot of real estate. >regards Graham Daniel. > >

Werner Terreblanche wrote: {Quote hidden}> > Steve > > Thanks for your reply. Inadvertantly, you also answered another question > that I had... namely "Is it possible to store WAV files directly in ROM and > play it back?" I have never been able to confirm how a WAV file is > encoded. Are you saying it is possibly to store the WAV file directly in > ROM and play it back just like that through a D->A converter? > > Rgds > Werner > > -- > Werner Terreblanche <wterrebSTOPspamspam_OUTplessey.co.za>Tel : (021) 7102251 (office > hours) > > > ---------- > > From: Steve Baldwin[SMTP:spamBeGonestevebSTOPspamEraseMEKCBBS.GEN.NZ] > > Sent: Tuesday, April 21, 1998 2:44 AM > > Subject: Re: Generating Digital Speech cheaply > > > > > Basically I want to make a very cheap voice playback device capable > > > 1. Is it a feasible way of doing this? > > > 2. What sampling rate would I need? (I was thinking 8KHz @ 8 bits > > > resolution) Is this good enough? > > > 3. Has anybody perhaps done something like this already? Was the sound > > > reasonably clear? > > > > I did this with an 8051 (not that it makes any difference) a while > > ago and it worked well. One of the Windows WAV file utilities lets > > you play with bit sizes and sample rates, so you can try it on your > > PC and see if it sounds good enough. It will show you the data size > > too, if you are planning on saving the data as a WAV file in your > > ROM which is about as easy as it gets. > > > > Steve. > > > > > > ====================================================== > > Steve Baldwin Electronic Product Design > > TLA Microsystems Ltd Microcontroller Specialists > > PO Box 15-680, New Lynn http://www.tla.co.nz > > Auckland, New Zealand ph +64 9 820-2221 > > email: KILLspamstevebspamBeGonetla.co.nz fax +64 9 820-1929 > > ====================================================== > > AD558 it's very good D/A to play voice! (Analog Devices) Miguel.

> still slightly too expensive for what I had in mind. EPROM IC's are very > inexpensive these days, and I thought that maybe if store my messages as raw > digitized speech on a large EPROM and play it back through a resistor ladder > network A/D and filter, I would be able to regenerate speech messages at a it definately works. 27256 will give you 8 seconds of speech at 8192 Hz sampling rate. Use CMOS binary counter for address and a decoder (like 138) to select a chip. Hook the data bus in parallel to the resistor ladder and off you go. For playpack, use 3 PIC pins to select a sample and then use fourth pin to ouput 8912 Hz square wave to advance the address counters. You probably want fifth pin for counter reset but if you are not using all decoder outputs then you may use one of these for reset. Of course, if youwant a peripherial for host system then it's probably better to do without PIC and just use parallel port to drive the thing, all you need instead of PIC is a sample rate clock. > 2. What sampling rate would I need? (I was thinking 8KHz @ 8 bits > resolution) Is this good enough? For speech, absolutely. Just make sure you filter out the sampling noise. > 3. Has anybody perhaps done something like this already? Was the sound > reasonably clear? about the same as phone. Actually sounded better because you usually use better device for playback. --- Read on for todays random tagline... God is real, unless declared integer.