Enhancement: Support Mozilla Deepspeech

[ghost]

I'm surprised I'm the first person to raise this one. :)

Mozilla have been building datasets and software for privacy-respecting speech recognition, over at DeepSpeech. It's cool that speech_recognition will allow abstraction over so many APIs, so that software can be built to treat the SR backend as a utility. It would be especially cool if this let users of all those programs written for Google Voice or whatever switch over to DeepSpeech, and remove another surveillance device from the world.

DeepSpeech now ship models in their releases channel, and Mycroft are switching their backend to DeepSpeech shortly, so it's getting traction and the kind of support that should rapidly improve it.

Thanks!

[Matthias84]

Beside the API call, the lib need an public API endpoint for different languages.

[bredmor]

Beside the API call, the lib need an public API endpoint for different languages.

DeepSpeech is an offline library (like Sphinx), not a web-based service.

[Matthias84]

There are API endpoints for 'as a service' pretrained models and voice detection. For example mycroft.ai

[MikeyBeez]

They have a websocket.

[reuben]

FWIW DeepSpeech's focus is on client-side, offline recognition. If this package works offline with PocketSphinx there's no reason it shouldn't be able to work with DeepSpeech. Similar constraints apply, namely needing to download models for languages you want to transcribe.

[norbusan]

Hi all,
here is my shot at supporting deepspeech in the same vein as pocketsphinx.
#454
This can be definitely improved by using the streaming api of deepspeech, but there is hardly any documentation about how to use it so I am hesitating.

[MikeyBeez]

I've given up on deepspeech. It isn't accurate enough. I'm using Julius for a wake word and google for everything else. Thank you all.