Support for Danish ASR hviske

[osos]

Would it be possible to support the Danish specific ASR hviske-model (https://huggingface.co/syvai/hviske-v5.3)?

In real-life test it is a lot better than whisper. I made a python server for it here (https://github.com/osos/hviske-stt-server) but would probably be better natively in audio.cpp.

[0xShug0]

Added to my ToDo list!

[0xShug0]

@osos would you be willing to share some audio WAV files for me to use as test cases?

[0xShug0]

@osos not sure whats going on, but running offical code produces garbage
combëëëëelijke题elleresch앉 시작 시작 passing ...
Since I can't have a good python baseline, I'll close the issues now.

[osos]

Danish-Wav-for-test.zip

Please find attached some Danish voice for testing. The wording is the same as the filenames.

[osos]

@osos not sure whats going on, but running offical code produces garbage combëëëëelijke题elleresch앉 시작 시작 passing ... Since I can't have a good python baseline, I'll close the issues now.

Did you try the one from my repo?

[0xShug0]

@osos I will try after the holidy.

[0xShug0]

@osos Done. stay tuned for release.

request	num_beams	python ms	cpp ms	speedup	parity
0	1	672.610	239.861	2.804x	pass
1	1	154.297	153.340	1.006x	pass
2	1	179.215	149.169	1.201x	pass
3	5	397.700	330.994	1.202x	pass

[0xShug0]

@osos Released. Please try and let me know if you run into any issues.