search

rishikksh20 / TalkNet2-pytorch

TalkNet 2: Non-Autoregressive Depth-Wise Separable Convolutional Model for Speech Synthesis with Explicit Pitch and Duration Prediction.
MIT License
85 stars 6 forks source link

Synthesized speech to fast #2

Open Saartjes opened 2 years ago

Saartjes commented 2 years ago

Hi there, I must say I’m impressed by the results Talknet is able to achieve with relatively short training. I have been playing around with it but keep running into a problem. The synthesized voice speaks very fast. Way faster than the input training data.

Is this something you have encountered as well? Do you know of any tricks that might help counter this issue?

Helaly96 commented 2 years ago

How were you able to get started with the repo?