[Feature] Upgrade SenseVoice-Small model from 2024-07-17 to latest (2025-09-09)

[iF2007]

Please consider updates the offline SenseVoice-Small model from the initial 2024 release to the 2025-09-09 enhanced version.

Key improvements include significantly higher accuracy for Cantonese and Code-switching (CN/EN) scenarios via the integration of the WenetSpeech-Yue dataset, refined INT8 quantization for improved stability on ONNX Runtime, and superior non-speech event filtering.

These updates enhance overall robustness in noisy environments while maintaining the model's signature low-latency Real-Time Factor (RTF).

Current model: https://huggingface.co/csukuangfj/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17
Updated model: https://huggingface.co/csukuangfj/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-int8-2025-09-09

This should be a drop-in replacement.

Also it would be nice if it was clearer for the user to see the exact version of the model.

[csukuangfj]

2025-09-09 is dedicated to Cantonese.

If you don't focus on Cantonese, please don't switch your model.

[iF2007]

I guess it's true that the new model is much more refined on Cantonese, but overall is also better?

[csukuangfj]

I guess it's true that the new model is much more refined on Cantonese, but overall is also better?

no, for Mandarin ，please use the old model.

[iF2007]

Why is new model tuned for Cantonese but "please use the old model" for Mandarin? Is there any proof or report for performance degradation on non-Cantonese languages? Just curious to know if this claim is true and understand why.