Why is the model trained with audio channels as 2? Although training data is LJSpeech.

[Ashigarg123]

Am I missing something?

[signofthefour]

@Ashigarg123 Hi, Sorry for late replying.
It is my bad for laziness, I reused the named of parameters as in PriorGrad. Here is why audio channels = 2.

First, FreGrad predicts Wavelet features instead of Waveform. Therefore, the output will be [low_freq, high_freq] instead of [waveform]. So that, the number of output channel is 2 (each for low and high separately) instead of 1 (mono-channel waveform).

Please discuss more if it remain ambiguous.