Arduino and LPC-10 speech synthesis

[Mr.G]

Does anyone have any direct experience of producing speech, particularly using LPC-10 with arduino. A friend is making something and doesn't want to use wav or mp3 files due to storage requirements.

LPC-10 if you don't know is similar to the Texas instruments Speak and spell from the 1980s

ref:

https://www.vocal.com/speech-coders/lpc-10/

Kind regards
Mr.G
Gerald

[Richard Ibbotson]

I worked a lot with LPC-10 around 35 years ago with the Texas Instruments software on the TMS320. I have not tried It on Arduino but there are many processors even without dedicated MAC instructions which can compete on performance these days.

 

My opinion of LPC-10 is that it was amazing that using quite basic modelling of the vocal tract and simple excitation it is possible to make intelligible speech at 2400 bits per sec. Some improvement in speech quality can be achieved by adding other simple parameters at data rates up to 9600bps. However, it got stuck there in terms of improving sound quality.

 

On the other hand, 8K samples per sec at 8 bits gives good quality speech. This raw 64,000 bps can then be compressed which then beats LPC-10 on quality and bit rate in the 16000 bps plus region.

 

Derivatives of LPC-10 using code excitation (CELP) and GSM type coding holds the middle ground but is much more processor intensive.

 

So, I would say you need to be clear on your objectives. While LPC-10 is not trivial it has quite reasonable small processor demands, likewise working down from 64000 bps by source coding is not to demanding. Good quality speech can be achieved in between but at the cost of much more processing power usually on a DSP device.

 

Interested to hear what you choose,

Richard

--
You received this message because you are subscribed to the Google Groups "rLab / Reading's Hackspace" group.
To unsubscribe from this group and stop receiving emails from it, send an email to reading-hacksp...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/reading-hackspace/CAHr2rBx2FGw6oHp-a%3DRw_mFXLMvuwU72e2H299SJ2G3kpBs4pQ%40mail.gmail.com.

[Mr.G]

Thanks Richard

Yes, I too interfaced some systems in the 1980s probably in 6502 or Z80 and I have forgotten nearly all of it.

.
.

We've found two programs that can generate the  code

Qbox that runs in Win3.1/Win98 from some French site http://furrtek.free.fr/index.php?a=speakandspell&ss=9&i=2. but I can't make it actually do anything useful

and

Bluewizzard on github

https://github.com/patrick99e99/BlueWizard

that runs on a Mac. So I build a virtual machine and the bluewizzard gui actually runs, but still little idea what to do with it.

So I'll probably just give my friend remote access to my VM and see if he can figure it out.

Kind regards

Mr.G
Gerald

To view this discussion on the web, visit https://groups.google.com/d/msgid/reading-hackspace/00b101d75740%24afe2aa70%240fa7ff50%24%40btinternet.com.

[Richard Ibbotson]

In terms of implementation on an Arduino. I don’t think an 8 bit Atmel will be sufficient. We used TMS32010 which was 16 bit 5MIPS with dedicated MAC instructions. Written in assembler.

I think an ARM based Arduino could work. Though you might consider  processing cost and memory cost  in your solution.

Richard

Sent via Richard's Phone

On 2 Jun 2021, at 01:30, Mr.G <gtom...@gmail.com> wrote:



Thanks Richard

Yes, I too interfaced some systems in the 1980s probably in 6502 or Z80 and I have forgotten nearly all of it.

.
.

We've found two programs that can generate the  code

Qbox that runs in Win3.1/Win98 from some French site http://furrtek.free.fr/index.php?a=speakandspell&ss=9&i=2. but I can't make it actually do anything useful

and

Bluewizzard on github

https://github.com/patrick99e99/BlueWizard

that runs on a Mac. So I build a virtual machine and the bluewizzard gui actually runs, but still little idea what to do with it.

So I'll probably just give my friend remote access to my VM and see if he can figure it out.

To view this discussion on the web, visit https://groups.google.com/d/msgid/reading-hackspace/CAHr2rBxJCH2zQdR3u1y4AVA_FuPsOKNQAtkTMk0q1-KoBCT8%3DA%40mail.gmail.com.

[Peter Wheat]

This might be a good place to start?  https://github.com/adafruit/Talkie

---

From: 'Richard Ibbotson' via rLab / Reading's Hackspace <reading-...@googlegroups.com>
Sent: 02 June 2021 09:57
To: reading-...@googlegroups.com <reading-...@googlegroups.com>
Subject: Re: [RDG-Hack] Arduino and LPC-10 speech synthesis

 

To view this discussion on the web, visit https://groups.google.com/d/msgid/reading-hackspace/26F8A7C5-85CA-478B-996B-546210B39B94%40btinternet.com.

[Mr.G]

Peter

Yes, that's what we're using those 1200 word libraries.

I found the link to Qboxpro as he said in future plans 4 years ago. I found the BlueWizard by accident. If I was a bit cleverer I could probably port it back to something Linux like.

Future plans

1. I'm working on an encoder for generating your own recordings (See demo). Its not producing results of the quality I would like, but things are improving. For now Qboxpro, an unsupported old Windows application, can produce Talkie compatible data streams.

Kind regards
Mr.G
Gerald

To view this discussion on the web, visit https://groups.google.com/d/msgid/reading-hackspace/AS8P191MB144804F842123B14F62F96ED893D9%40AS8P191MB1448.EURP191.PROD.OUTLOOK.COM.

[Peter Wheat]

Perhaps try contacting pat...@collinatorstudios.com to see if he is aware of a Linux port?

---

From: reading-...@googlegroups.com <reading-...@googlegroups.com> on behalf of Mr.G <gtom...@gmail.com>
Sent: 02 June 2021 13:06

To: rLab / Reading's Hackspace <reading-...@googlegroups.com>

To view this discussion on the web, visit https://groups.google.com/d/msgid/reading-hackspace/CAHr2rBx028-JJMWbT2%2BNC%2Bewa9uuLnUeZpnicHAJA7%2BrCAV1-w%40mail.gmail.com.

[mikethebee]

I've played with the open source Alexa code for such as the Seeed Respeaker 2 pHat RaspberryPi, this is mostly about voice recognition rather than reproduction, but can talk back. Also, I do still have an IC that did phonemes from text via a parallel printer port, but I would need to look up the type spec. if you are interested.