JorisCos

/

ConvTasNet_Libri3Mix_sepnoisy_16k

like 1

Is this model bugged in some way? It looks like it expects 8k audio even though the model has 16k in it's name

from asteroid.models import BaseModel
voice_filter = BaseModel.from_pretrained("JorisCos/ConvTasNet_Libri3Mix_sepnoisy_16k")
self.voice_filter.separate(audio_path)

Gives me this error
RuntimeError: Received a signal with a sampling rate of 16000Hz for a model of 8000Hz. You can pass `resample=True` to resample automatically.

I also got the same from the version 2 sepclean, JorisCos/ConvTasNet_Libri2Mix_sepclean_16k
and version 2 noisy JorisCos/ConvTasNet_Libri2Mix_sepnoisy_16k

Idk if other models have the same issue. I'd rather not use resample=True b/c the audio quality is soooo bad