Google's TTS team has a new text-to-speech engine for us using LSTM (Long short-term-memory) Recurrent Neural Networks. Sizes are pretty similar, let's try to include the same set of 13 voices as part of Chrome OS natively.
@kathrelkeld and @lpalmaro we've updated the TTS engine in M53 and I've tested it on Canary; could you please help sanity-check that nothing has regressed?
Would be worth checking:
* Each supported language (de, en, es, fr, hi, id, it, ko, np, pt-BR)
* Each architecture (x86, x86-64, ARM)
* Low-end devices (the new engine may require a bit more cpu)
We can poke at some of this during M53 FullRelease (early July-ish), but a detailed-scope check on various languages is out of our wheelhouse.
Laura, shall we kick off another LQA pass once M53 is in beta?
Any time before stable should be fine. If we find a showstopper we can just revert back to the previous engine safely. A detailed look at the languages isn't necessary. Thanks!
Comment 1 by bugdroid1@chromium.org
, Jun 12 2016