Some Beginner Questions

[ok_ok]

Hello Guys,

im currently trying to Evaluate some of the “most accurate” STT-Models and i have some Questions about DeepSpeech / Common Voice.

1. I guess DeepSpeech is trained on the full Common Voice (English) Dataset. Would there be any reason for me to Download the full Dataset, and try to Augment the Data by adding Noise and then Fine-Tune the Model with that Data?
   Or should i only try to Fine-Tune the Model if i have new data?

2. which datasets is DeepSpeech trained on?

3. If would train a new Model in another Language, would i still do Transfer-Learning or would i have to train the Model from Scratch?

4. Are there any resources you guys would recommend to a beginner, who wants to start with DeepSpeech?

5. Is there maybe a List with Pretrained-Models including some Scores so that i know on which Model i should build on?

Thanks Guys

[lissyx ((slow to reply) [NOT PROVIDING SUPPORT])]

@ok_ok

I guess DeepSpeech is trained on the full Common Voice (English) Dataset.

This is documented in the release description: https://github.com/mozilla/DeepSpeech/releases/tag/v0.9.3

@ok_ok

which datasets is DeepSpeech trained on?

same link

@ok_ok

If would train a new Model in another Language, would i still do Transfer-Learning or would i have to train the Model from Scratch?

That depends on what you want to achieve and what data you have

@ok_ok

Are there any resources you guys would recommend to a beginner, who wants to start with DeepSpeech?

Documentation? Playbook? https://deepspeech.readthedocs.io/ https://mozilla.github.io/deepspeech-playbook/

@ok_ok

Is there maybe a List with Pretrained-Models including some Scores so that i know on which Model i should build on?

It’s already listed in a thread on discourse.