Topic awaiting preservation: JS1K Speech Synthesizer

Ahoy! Long time no see... I drop by every now and then to see how's the asylum going.

You all know I'm crazy about JavaScript and size optimization. You probably also heard the JS1k contests. This contest gives you a shim HTML5 page with a canvas and 3 global variables to access the body, the canvas and its 2D context, and you have 1024 bytes of JavaScript to kick ass. No external resources, external API call, nada allowed!

JS1K#4 closed recently and the results are expected at JSConf US, on April 2 & 3.
This time the contest collided with our long awaited vacations but I definitely wanted to enter. The optional theme was "love", also Audio was game this time. Of course there was tons of flying hearts and flowers and yaddi blah, and some entries featured a little melody and simple sound effects.

But, no heavy 3D, FPS game, particle systems, flying hearts and flowers from me this time. Audio was ON and I had to make a Speech Synthesizer in 1K of JavaScript.

It's a simple formant based synthesizer that speaks whole sentences AND as you type. This puppy supports 25 sounds/phonemes.

I hope this zero polys and zero shaders* entry will catch people's eye^Wears.



(Edited by poi on 03-16-2012 20:44)

would love to try it - only I get a 404 on your link?

I grabbed this from his twitter feed

"made it to the JavaScript Weekly newsletter. Next stop:"

but I can't get it to work.... but what do I know. -_Q


http://js1k.com/2012-love/demo/1274

___________________________________________________________________________
"It doesn?t matter to business, if cannibalism were legal, we would all be in cans."

Go me! I gave the wrong URL. Here's the correct URL of the JS1k Speech Synthesizer

I could only test on Windows. It works perfectly fine in Opera, Firefox and Chrome.
Safari was throwing an exception on new Audio(...) can you say spec violation ?

The demo page generates and speaks the default sentence right away. You can edit the giant text input featuring the default sentence. The synth will play a sound for each keypress and will generate and speak the whole text upon pressing ENTER.





(Edited by poi on 03-17-2012 18:50)

amazing. Even though the focus is on the synthesizer, not the speech.

Can't quite say that the 1k limit makes the code more readable .
Would you elaborate about the method you use? The tuples per character look so very short
for coding waveforms..., but then I have no clue about this at all.

Thanks.
What do you mean by "The tuples per character look so very short for coding waveforms" ?

The synth + player is basically an optimized version of my Music Synth in 140bytes + player: The sounds are generated as an 8bits mono 8Khz WAVE PCM file, fed to a new Audio element ( as a data: URI ) that I play right away. Being 8bits mono, means that only one byte is needed per sample of the sound, and being 8Khz means that 8000 bytes are needed to make 1s of sound. Pretty basic math

The formant part, I ported to JS and optimized Tiny Speech Synth by Stepanov Andrey.

To fit in 1k I had to use 2 formant filters instead of 3 thus decreasing the size of the data and the quality of the speech. Also I ditched support for plosive ( overlapping ) sounds thus greatly simplifying the synth and allowing me to synthesize the speech in a single pass.

Hope that clarifies a few things.



(Edited by poi on 03-19-2012 07:38)

It's alive!!! Good stuff po1

Very cool once again poi

Thanks.

I just tweaked the code, switched to my experimental packer and made the Audio element visible, thus adding support for volume setting/pause/stop/replay and sound export in over 100bytes less than the initial release.