Does not support long audio inference？ about sherpa-onnx

[csukuangfj]

Does this mean that if I want to deploy the same model both online and offline, I need to export two separate model files?

Yes.

@csukuangfj Is there a corresponding guide for exporting the model separately for online and offline models? Thank you very much for your response.

Please see our icefall doc for model exporting.

[csukuangfj]

Whisper supports at most 30 seconds.

To support longer audios, we have to break then into chunks, each of which should be less than 30 seconds long.

We provide VAD + ASR.
You can try silero VAD + Whisper tiny.en Android APK provided by us.
Please see https://k2-fsa.github.io/sherpa/onnx/vad/apk-asr-cn.html

(Specifically, the APK for whisper is https://hf-mirror.com/csukuangfj/sherpa-onnx-apk/blob/main/vad-asr/sherpa-onnx-1.9.25-x86_64-vad_asr-en-whisper_tiny.apk)

Currently, we have no plan to make test.py support longer audios.

[MrRace]

Whisper supports at most 30 seconds.

To support longer audios, we have to break then into chunks, each of which should be less than 30 seconds long.

We provide VAD + ASR. You can try silero VAD + Whisper tiny.en Android APK provided by us. Please see https://k2-fsa.github.io/sherpa/onnx/vad/apk-asr-cn.html

(Specifically, the APK for whisper is https://hf-mirror.com/csukuangfj/sherpa-onnx-apk/blob/main/vad-asr/sherpa-onnx-1.9.25-x86_64-vad_asr-en-whisper_tiny.apk)

Currently, we have no plan to make test.py support longer audios.

@csukuangfj
Wow, the provided APK looks great. Is there a detailed guide available for compiling these APKs? I may need to fine-tune my own model. Thanks a lot!

[csukuangfj]

Please see our doc.

https://k2-fsa.github.io/sherpa/onnx/android/index.html

[MrRace]

@csukuangfj
If I want to deploy Whisper in online mode to a mobile phone, it seems that the current guide only supports offline mode?
So, if I want to deploy the Whisper model with online mode to a mobile phone, how should I modify the code that I can add a new model? Thanks! Additionally, I noticed that some models in the download list have "stream" in their names while others do not. Does this mean that if I want to deploy the same model both online and offline, I need to export two separate model files?

[csukuangfj]

If I want to deploy Whisper in online mode to a mobile phone, it seems that the current guide only supports offline mode?
So, if I want to deploy the Whisper model with online mode to a mobile phone, how should I modify the code that I can add a new model?

We also want to do that. Please tell us if you find a framework supporting that.

---

Does this mean that if I want to deploy the same model both online and offline, I need to export two separate model files?

Yes.

[MrRace]

Does this mean that if I want to deploy the same model both online and offline, I need to export two separate model files?

Yes.

@csukuangfj Is there a corresponding guide for exporting the model separately for online and offline models? Thank you very much for your response.