README.md

metadata

license: mit
language:
  - ja
size_categories:
  - 1K<n<10K

TTS Training Detail

Size-Category - Large(500-5000):1230 files

Audio Detail

• 22050hz mono wav - for modern tts (original are 32khz mono mp3)
• vadded 200ms
• Volume-level:OK - no need to gain.
• Noise-level:OK - no need to denoise

Creating Detail

• Based 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)

Don't attempt to identify the identity of the speaker in the voices.

I agree to download original public domain dataset.