ASR

The Next-gen Kaldi not only provides solutions for training speech recognition models and deployment, but also releases a large number of pre-trained models and corresponding demo programs.

Huggingface space¶

The most direct and convenient way to experience the Next-gen Kaldi is to visit our provided Huggingface space with a browser, which currently supports the experience of dozens of models in languages such as Chinese, English, Chinese-English, Chinese-English-Cantonese, Cantonese, Tibetan, Arabic, German, French, and Russian.

Huggingface Space。

Webassembly¶

The Next-gen Kaldi also offers webassembly support, which allows you run the models totally in the browser and no server is needed. Here are several pre-compiled models in the following table. If you want to package your own models using webassembly, plsease refer to sherpa-onnx docs and sherpa-ncnn docs

Subtitle extraction¶

The subtitle extraction is a small tool we made based on speech recognition. Users can upload a video to generate the corresponding subtitle file. Huggingface space (support Chinese, English, Chinese-English, and Russian) address .

Videos¶

Note: All videos are in Chinese!

In order to make it easier for you to see the demo quickly, we have also made a lot of videos, see Bilibili for details.

apk & exe¶

We also offer a number of compiled android APK and Windows executables.

Comments