NES Speech Synthesizer

Hello all,

I'm working on a Speech Synthesizer demo for the NES. This is a project I've wanted to work on for a very long time. It's about half finished, but it's taking input and talking. I made a lot of effort to get it to look like a ~1980s computer screen.

Currently, all the DMC samples are at sample rate "a", in order to fit it into an NROM size ROM. The sound quality is not great, but I don't think I can fit in a faster sample rate. Maybe I'll try "b" and see if I can move things around, but it will be a pain, there are 32-33 samples.

I'm not ready to release the .nes ROM, but here's a YouTube video.

https://youtu.be/kl3NMzgF1eY

Interesting!

Nice! It sounds a lot like the one I made a while back. I never released the ROM because I never properly finished the editor for it, the saving and loading, and backspace worked in the audio but not the display. If you want to hear it, it's used on the last track (#183) of this NSF:
http://membler-industries.com/nes/Memblers_-_Complete_Chipography_1_of_3_(Original_Songs).zip

The fun thing about not using the highest DPCM speed, is that you can make the samples higher pitched, heheh. My editor let you set the speed setting for each phoneme as you entered it, not greatly useful but can be kinda funny.

I was honestly expecting something like the Intellivoice Voice Synthesis Module based on the title (presumably plugged into the expansion port)

That or a port of SAM (Software Automatic Mouth).

I'd bet MMC3 can do an actual diphone synthesizer.

Wow!

This is really impressive!

Why just NROM?

Maybe in the future I will try better samples / better mappers. I wanted to keep it as simple as possible, with no bank switching.

It's a cool program.
This is one of those programs that would benefit from the Family Keyboard.

It's no coincidence, that the folder that holds the source code has SAM in the title. I didn't know that anyone else was aware of SAM.

By the way, it's 95% done. I just want to add some sample text options, and I will post it.

Really? I even knew what that was, I just didn't know what the acronym was (or that it even meant something and wasn't just a name for it).

Then again, I know a bunch of random crap from being here.

And she's done, here's the link...

http://dl.dropboxusercontent.com/s/3x5c ... alkNES.zip


Please let me know if you find any bugs. Thanks.

I think it needs help... I had no chance at understanding the first and third samples (hate to talk like this, but the first sample sounded like it said "suck dick" multiple tikes) I'd have no chance at understanding the Hamlet passage if we hadn't had to memorize it for school. It's certainly a start though, but I wouldn't say it's complete unless you're just tired of it.

Well, Speech Synthesizers that sound better use more samples, and higher quality samples. Maybe I'll play around with it in the future, to try to improve the sound (maybe use a different mapper), but I'm pretty happy with the results so far.

I'd rather get back to work on my other game, for now.

And some don't use any samples but instead formant synthesis using modulated sine waves, like SAM.

Many years ago...I tried to generate speech using combinations of sine waves...some with pitch changes... at different frequencies. Etc.

In the end, I couldn't make a single vowel sound that reminds me of actual speech.

Samples l understand. It was so much easier. But, this project was for fun. ...Spending months trying to figure out a way to make Square channels make vowel sounds... Is not fun, and maybe not even possible.

This is really neat! Reminds me of Johnny 5, heh (before he became sentient).

(in response to synthesis using square waves...) I wonder if it'd be possible by updating the APU faster than once per frame and using a technique something like this: https://youtu.be/muCPjK4nGY4 maybe not since you might have to use many more channels to get the desired effect...but then, you can modify the timbre somewhat, and the noise channel could help with consonant sounds perhaps. Not that I'd ever try this myself, it sounds quite daunting indeed, and probably a lot of precalculation outside of the NES itself would be required.

I made a new version...
-edited each WAV file, louder, EQed, slightly increased the pitch, edited some for length
-resampled every one to DMC rate 'D' (I was able to cram them all into an NROM space)

Is this a bit more understandable?

http://dl.dropboxusercontent.com/s/58d3 ... lkNES4.zip

Wouldn't that be like voice synthesis using CSM on some Yamaha synth chips? I know of a couple of NSFs that use sines on the Namco163 and a few files on the sharp X1 that accomplish this but they're all anime openings and terrible memes. It's a very impressive concept that few companies ever put into practice, with the only one I can think of being Game Arts using it in Silpheed for the PC-88.