How can i process Chinese Mandarin speech recognition

I have read the docs of version 0.7.4, but i still can’t get a full vision in chinese Mandarin speech recognition.
Usually, in other aticles, they offen don’t use unicode codepoint and do two things more.
The first is dividing sentence into phrases in language model,
and the second is using chinese pinyin(chinese phoneme) as a media between sound and character.

So my question is how can i process Chinese Mandarin speech recognition using our model more effectively.

Great to have you here. Why don’t you start by setting up an English training so you get to know DeepSpeech as a software. Then you can switch over to Mandarin.

@reuben has been training a Mandarin model for some time, don’t know whether there is a repo or other resources you could use?

1 Like

@othiele For those people who knows how DeepSpeech works, do you know where to find trained Chinese voice models and scorers?
Appreciate your comments!

Chinese seems to be hard to find for Mozilla DeepSpeech and there were some questions about it. Maybe search the forum and ask others.

But as we are an open community, there seems to be one in a similar, but different, repo. If it works, please let us know.

To be honest, I was trying it last night. But the repo didn’t come with a trained model. It comes with a Chinese voice dataset which is more than 150GB. I don’t think my homemade workstation can load that work… But it works anyway. The only thing is that no public trained model for now… I’m looking for…

I meant this model section, they are about 800 MB.

Thank you. I downloaded it.
the compressed file is like this:


Do you know how to use it? https://github.com/PaddlePaddle/DeepSpeech#speech-model-released
File params.pdparams is 800MB.
I didn’t find how to use the deepspeech engine with the trained model… I have downloaded the docker image from PaddlePaddle…
Is it like this? deepspeech --model params.pdparams --scorer some.scorer --audio myaudio.wav?

Sorry, we don’t support the competition :slight_smile:

This project has nothing to do with the other, therefore it is hard to say much. But they have a file called infer.py