SpeechGrammarList + Alternative transcripts

[fabswt]

Hi everyone!

I’m new to DeepSpeech (it was surprisingly easy to install, thanks for the proper documentation), looking for an alternative to Google Chrome’s speech recognition API.

A few noobie questions I couldn’t find an answer to here or in the FAQ:

• Does DeepSpeech support SpeechGrammarList to provide a list of keywords to be recognized? It seems the answer is ‘no’, but then are there any effort in this direction? SpeechGrammarList is part of the Web API draft specification so I expected people to be working on it.
• How do I get alternative transcriptions, instead of just the one result?
• What is the current English model based on (deepspeech-0.4.1-models.tar.gz)? (CommonVoice data? LibriSpeech data? I didn’t see any credits in the model files.)

Thank you for your time.

Best –
Fabien

EDIT: for context, I’m trying to recreate this MDN phrase matcher demo using DeepSpeech.

[lissyx]

How do I get alternative transcriptions , instead of just the one result?

The API does not provide that for now.

Does DeepSpeech support SpeechGrammarList to provide a list of keywords to be recognized? It seems the answer is ‘no’, but then are there any effort in this direction? SpeechGrammarList is part of the Web API draft specification so I expected people to be working on it.

I’m unsure this should be handled in DeepSpeech, likely more in a WebSpeech API implementation that relies on DeepSpeech. I mean, at some point, it likely is just a matter of solving the issue for multiple language models, and then tuning the weights?

What is the current English model based on ( deepspeech-0.4.1-models.tar.gz )? (CommonVoice data? LibriSpeech data? I didn’t see any credits in the model files.)

It’s documented in the release: https://github.com/mozilla/DeepSpeech/releases/tag/v0.4.1