metadata
license: mit
datasets:
- mozilla-foundation/common_voice_16_1
size_categories:
- 1K<n<10K
Don't attempt to identify the identity of the speaker in the voices.
I agree to download original public domain dataset.
Audio Info
- 22050hz mono wav - for modern tts (original are 32khz mono mp3)
- vadded 250ms (little bit long)
- Size-class:Huge(9522 files) - it's take long time to train,but possible to create high-quality tts from scratch.
- Volume-level:OK - no need to gain.
- Noise-level:OK - no need to denoise
Detail
- part of common_voice_16_1 ja
- audios are vadded with silero-vad
- text are cleaned simply by openjtalk.(this mean possible contain miss reading and miss converting)