alphacep / vosk-api

Offline speech recognition API for Android, iOS, Raspberry Pi and servers with Python, Java, C# and Node
Apache License 2.0
8.12k stars 1.12k forks source link

Please add vosk-model-small-en-us ver. 0.22 #1644

Open Slavczays opened 1 month ago

Slavczays commented 1 month ago

I noticed that small english 0.15 version has rather poor performance (compared to whisper tiny/base). Could you please add model vosk-model-small-en-us but 0.22 version (for portable devices, android, RPi). I don't mean vosk-model-en-us-0.22-lgraph whose performance is also very average. I would like to point out, however, in case of Polish language your model vosk-model-small-pl-0.22 is simply fantastic, in my opinion much better than whisper tiny/small and works much faster. So I guess english 0.22 small version would be also much better, unfortunately base version (vosk-model-en-us-0.22) is too big (1.8G) for portable devices.

nshmyrev commented 3 weeks ago

Thank you! We are working on new versions, but they are not going to be available soon, we have too small resources unfortunately. Meanwhile you can use some modern recognizers like sherpa-onnx probably. You still need good models for them.

Slavczays commented 3 weeks ago

Thanks. Then I'll wait patiently;)