kan-bayashi / ParallelWaveGAN

Unofficial Parallel WaveGAN (+ MelGAN & Multi-band MelGAN & HiFi-GAN & StyleMelGAN) with Pytorch
https://kan-bayashi.github.io/ParallelWaveGAN/
MIT License
1.54k stars 339 forks source link

A Length Mismatch in Recipes that use segmentation #412

Closed South-Twilight closed 1 year ago

South-Twilight commented 1 year ago

Hi, thanks for the great repo. When I run recipes that need to segment wav, just like OFUTON, CSD, ect., there's an error below:

Stage 1: Feature extraction
Feature extraction start. See the progress via dump/train/raw/preprocessing.*.log.
Feature extraction start. See the progress via dump/dev/raw/preprocessing.*.log.
Feature extraction start. See the progress via dump/eval/raw/preprocessing.*.log.
ERROR: wav.scp and segments has different #lines (47 vs 547).
ERROR: wav.scp and segments has different #lines (5 vs 70).
ERROR: wav.scp and segments has different #lines (5 vs 58).

It's a length mismatch between length before segment in wav.scp and length after segment in segments. I locate this error message at utils/make_subset_data.sh L40-L54.

How can I solve this problem? If I want to make wav.scp be the length after segmentation, what shuold I do?

kan-bayashi commented 1 year ago

Sorry for late reply. This is my mistake. I will fix the issue.

kan-bayashi commented 1 year ago

Fixed in #413. Could you try again?

South-Twilight commented 1 year ago

Fixed in #413. Could you try again?

It's ok to run now. Thanks.