This demo implements offline speech recognition and speaker identification for mobile applications using Kaldi and Vosk libraries.
Simply import the project into Android Studio and run. It will listen for the audio and dump the transcription.
To use this library in your application simply modify the demo according to your needs - add kaldi-android aar to dependencies, update the model and modify java UI code accodring to your needs.
This is just a demo project, the main setup to compile vosk-android library AAR is available at vosk-api. Check compilation instructions there as well as development plans.
Models for different languages (English, Chinese, Russian) are available in
Models section. To use the model unpack it into
kaldi-android-demo/models/src/main/assets/sync/model-android
. More languages gonna be ready soon.
To run on android model has to be sufficiently small, we recommend to check model sizes in the demo to figure out what should be the size of the model. If you want to update the grammar or the acoustic model, check vosk-api documentation.