I Made a 10¢ Mcu Talk

Interesting, though before clicking I thought the headline might be referring to a very poorly paid presentation about Marvel movies.

This is great. Missed opportunity for a low-pass RC filter on the speaker circuit - if you know you're driving an 8kHz sample rate, you can design your filter with that cutoff, and it'll sound way better (it'll get rid of the buzzy quality).

You should be able to do it all on-device, check out SAM, the Software Automatic Mouth. The actual data in the *_tabs files:

https://github.com/ctoth/SAM/tree/master/src

First I thought you made a lecture on MCUs which was available for viewing in exchange for $0.10

Then I thought you made a lecture on MCUs where the device was available for purchase generally for $0.10.

Then I thought with an MCU valued at $0.10 you generated speech

English... sigh

These CH32 mikrocontrollers are great and dirt cheap. I've build a small DC motor controller with them to control a toy robot: https://github.com/h0lad/MiniSpeedController

The bigger ones have PHYs for USB HS, USB-C (5Gbps) and 10/100M Ethernet integrated (!). And their development environment (Mounriver Studio) isn't too bad - I didn't had the immediate urge to port everything to CMake/VSCode.

But they need some kind of pin planning tool. It's awful to use the datasheet and find the correct pin functionalities and their mutual exclusions... STM32 mastered this with their STM32CubeIDE tool: select a feature (like USART1) and the right pins light up - alternate pins are easy to locate.

They also should clean up their license mess on OpenWCH (their GitHub page). Lots (all?) of their HALs are Opensource - but the right version with right SPDX tags are often a bit hidden.

Even easier is the 1-bit DPCM codec used on the NES. Go up or down one step each sample depending on the next bit. You can't get a square wave out of it, only a triangle, but it has a nostalgic quality.

Programmers made 8 bit micros talk in the 1970s, using nothing but square wave pulses sent to the speaker.

It's probably worth mentioning the 2400bps (300 bytes per second) LPC10 codec built into SoX. If you have SoX installed, try

    rec -t lpc10 speech.lpc

and then speaking into your microphone for ten or fifteen seconds before you ^C it. Then play it back with

    play speech.lpc

It will sound very robotic but pretty comprehensible, at least with an adult male voice in English, and it preserves a lot of the prosody and enunciation that is so hard to get out of speech-synthesis packages.

12KiB of data at 300 bytes per second would be 41 seconds of recorded speech.

Decoding the LPC10 data on the CH32V003 might be tricky. On amd64, running `make CFLAGS=-Os` followed by `ld -r -o tmp.o *.o` inside sox-14.4.2+git20190427/lpc10 yields a tmp.o with 25243 bytes of text (including .rodata, etc.) and 356 bytes of data. I'm not optimistic that RISC-V would compress that to fit inside the CH32's flash. And I find the code in that directory inscrutable; it's Fortran that's been compiled to C.

Still, it seems plausible that you could massage the LPC10 data into a format that something like Talkie would understand.

Could hang an i2c flash chip off that thing for more storage and still have enough IO pins for serial coms and a spare IO pin.

Have you tried just savimg the zero crossing times?

From distant memory you got quite a bit more compaction.

> I considered a few encoding options for compressing the audio.

The presentation of this part seems extremely padded out to me, ironically enough.

I saw another audio project on the same microcontroller (family) posted a few days ago: ModPlayRISCV It plays a tracker MOD. using PWM with a low-pass filter. It resamples/scales all samples at varying rate/volume into a ring buffer which gets fed to the PWM comparator by DMA.

https://github.com/cpldcpu/ModPlayRISCV

And most code is in C++, while I was expecting straight Assembly, pleasently surprised.

Nice work. Especially referencing the TI prior art of the Speak and Spell. This kind of synthesis was quite prevalent in the early 80s - school BBC Micros had a ROM which let you "*SAY" a phrase. Classic Macs had MacinTalk.

Another codec which might be interesting to try but is considerably more complicated is AMR, from GSM: https://en.wikipedia.org/wiki/Adaptive_Multi-Rate_audio_code...

Can you export the schematic and Gerber files to a PDF file? A lot of open source projects do this, and it makes it much easier to tell what's going on, with software pretty much everyone already has on their computer.

It would need to be ported, but the Talkie library for the AVR / STM / SAMD / ESP, with its roots dating back to the TI speak and spell toy, gets a phenome engine with a good vocabulary into less than 8k. It’s not musical though lol.

The pwm on the CH32v003 is pretty similar to the STM32 implementation, so porting might not take much.

It would be really cool to have phenome/text based vocabulary like talkie on that little chip! Since it uses text/phenomes it can have a large vocabulary for such a tiny chip. It would be possible to have about 1500 words in an 8k dictionary, 2k for code, and 4K for the phenome engine and still fit in the 16k of flash it comes with.

Incidentally, there is another riscV from WCH that also features BLE, 200K+ of flash, and 18K of RAM in an ESSOP-10 package (same size as the SOP8 but only 4 GPIO). It’s around $0.41 in Q1. The vocabulary with that would be 20k plus words with 100k left for code lol.

It’s just nuts what a dollar will get you these days in that space.