Open darxkies opened 3 weeks ago
Thanks for reporting this, for the first error I just pushed #2442 which should get things to work on cpu. I've also added support for the mini model via --which mini-v1
.
For incomplete generation, you may want to try increasing the temperature with --temperature 0.6
or try other values. I've also found the python version to be a bit unstable on low temperatures so not sure how much of the issue comes from candle (I haven't been able to find the usual TTS evaluation metrics like word error rate to get a sense of the quality to expect from this TTS so it's hard to know what to expect here).
When running the Parler-TTS example commands it ends with the following error:
It works with CUDA though, but only as long as the default prompt is not changed. If it is changed the generated audio is incomplete.
It also crashes when choosing the mini model with the error: