search

TensorSpeech / TensorFlowTTS

:stuck_out_tongue_closed_eyes: TensorFlowTTS: Real-Time State-of-the-art Speech Synthesis for Tensorflow 2 (supported including English, French, Korean, Chinese, German and Easy to adapt for other languages)
https://tensorspeech.github.io/TensorFlowTTS/
Apache License 2.0
3.85k stars 814 forks source link

WebGL / in-browser synthetisation opportunity? #451

Closed miohtama closed 3 years ago

miohtama commented 3 years ago

The latest WebGL supports TensorFlow:

https://www.tensorflow.org/js/guide/platform_environment

My question is that are there any technical limitations, like model sizes or specific used TensorFlow features, which would prevent running TensorFlowTTS within a web browser using WebAssembly and WebGL. As this would open a door dor dynamic dubbing and would be a huge accessibility win in many uses. Are there any other considerations one should think when embedding TensowFlowTTS into a web page?

Myself, I consider CDN cached 100 MB - 500 MB download size for the model as the reasonable expectation what users would be willing to download to get the text-to-speech within a web page or web app.

dathudeptrai commented 3 years ago

i never try to use webGL :)))

StoneCypher commented 3 years ago

Myself, I consider CDN cached 100 MB - 500 MB download size for the model as the reasonable expectation

so, ... half a cent per page view?

This isn't realistic.

miohtama commented 3 years ago

There are open source-oriented CDNs like KeyCDN and others. These CDN services are happy to absorb the distribution cost if the service for the public good in the exchange of brand and promotion value. The internal price of delivery is way lower than half a cent per user, so it is not unrealistic. You do not need to load the set for every page view.

stale[bot] commented 3 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.