Necessary preprocessing for inference wav data

[takan1]

Hi, Thanks for the great work! I'm trying to test the inference part with my own wav file but the output quality is less than I expected and I'm suspecting it's due to the input file.
Could you give me some instruction for how to preprocess the input source/target wav?

[OlaWod]

For source wav, it is better to denoise it if it is too noisy, normalize it if the volume is too loud or too low; for target wav, it is better to trim its silent segments.

[takan1]

@OlaWod Thanks a lot for the answer! Understood. I'll try the above. Also, would there be a preferred sampling rate for both source and target wav?

[OlaWod]

16kHz

[takan1]

Great, thanks for your response :)

[Crazy-Duck]

For the target wav, do you have recommendations as to the length? Does performance of the model improve/deteriorate with longer target wavs? If so, where's the sweet spot?

[OlaWod]

haven't explored yet, but I think longer target wav might contain longer silent, which can deteriorate the performance. I think the sweet spot is a target wav that is not too short and contain almost no silent.