Closed epii2zero closed 9 months ago
Thank you for nice work!
I have a question about encoder loss In paper they used MSE for encoder loss (section 3.1)
but in this implementation, use different loss https://github.com/p0p4k/pflowtts_pytorch/blob/297166f518263b2822f52918f86f7d27fa72fe8b/pflow/models/pflow_tts.py#L179C9-L180C74
I think this loss came from Matcha-TTS codes Is this intentional?
Thank you!
Essentially it is the same loss, scaled. (at a constant variance of 1, gaussian nle becomes mse)
Oh I see, thank you for reply
Thank you for nice work!
I have a question about encoder loss In paper they used MSE for encoder loss (section 3.1)
but in this implementation, use different loss https://github.com/p0p4k/pflowtts_pytorch/blob/297166f518263b2822f52918f86f7d27fa72fe8b/pflow/models/pflow_tts.py#L179C9-L180C74
I think this loss came from Matcha-TTS codes Is this intentional?
Thank you!