MorganCZY

about train_tacotron loss definition

4

In your implementation of tacotron, mel_outputs of decoder are fed into the postnet(a cbhg) and a linear layer, then getting a new outputs with the same dims of mel_outputs. You...

About Loss Curve and timbre of converted voices

4

Hi, After listening carefully to converted voices, I found their timbre is not too much like the target's. Then I reviewed all the loss definitions and their training curves. As...

about final loss?

16

Could you post your training loss curves with LJSpeech dataset? After 3k epochs, synthesized waves by my trained model are of poor quality compared to the released model "linda_johnson.pt". I...

I notice the original implementaion of fastspeech(integrated in ESPNet) adopts log domain to calculate the duration loss, which means target duration is first token the logarithm. In your version, the...

about Chinese and English dic

1

Could you briefly introduce the function and content of files in "dic"(*.conf, *.macron, *.table)? Besides, to synthesize Chinese(or English) singing voice, is the file "chinese.table"(or "english.table") enough for this task?...

the trained model generates different wavs with the same text and reference audio

15

When doing tests, I found each time I ran the synthesize.py(with the same text and reference audio), I got different results(namely different syntheized wavs). After looking up the code, I...