alibabasglab

/

AV_MossFormer2_TSE_16K

like 1

I figure it out on app.py .
def fn_clearvoice_ss(input_wav):
myClearVoice = ClearVoice(task='speech_separation', model_names=['MossFormer2_SS_16K'])
output_wav_dict = myClearVoice(input_path=input_wav, online_write=False)
if isinstance(output_wav_dict, dict):
key = next(iter(output_wav_dict))
output_wav_list = output_wav_dict[key]
output_wav_s1 = output_wav_list[0]
output_wav_s2 = output_wav_list[1]
else:
output_wav_list = output_wav_dict
output_wav_s1 = output_wav_list[0]
output_wav_s2 = output_wav_list[1]
sf.write('separated_s1.wav', output_wav_s1[0,:], 16000)
sf.write('separated_s2.wav', output_wav_s2[0,:], 16000)
return "separated_s1.wav", "separated_s2.wav"

Hey, were you able to use it? my use case is also based on audio visual information.

yes i got the results. i have written the function i use above. but the perdormance was not good. it confuses the voices . the person one becomes person 2 for example. and in my usecase I wanted to seperate multiple people in a talk. which this is not helpful.