README.md

metadata

language:
  - en
  - ja
pretty_name: Kit Lemonfoot's Voice Datasets
task_categories:
  - audio-to-audio
  - text-to-speech
tags:
  - audio
  - speech
  - vtuber
  - youtube
  - voice

Kit Lemonfoot's Voice Datasets

This repository aims to house every dataset used in my AI voice models. Please credit me if you use these datasets.

This includes the following:

• Every dataset used in my RVC models
• Every dataset used in my Style BertVITS2 models
• Every dataset used in my GPT-SoVITS models
• Multiple datasets submitted to 15.ai (RIP)

Reference

Hand-transcribed datasets were built for 15.ai and as such are in the LJSpeech transcription format.

SPEAKERNAME/filename1.wav|transcription
SPEAKERNAME/filename2.wav|transcription
SPEAKERNAME/filename3.wav|transcription

Whisper-transcribed datasets are in the VITS transcription format as seen in BertVITS2 and its' contemporaries.

filename1.wav|SPEAKERNAME|LANGUAGE|transcription
filename2.wav|SPEAKERNAME|LANGUAGE|transcription
filename3.wav|SPEAKERNAME|LANGUAGE|transcription

While some non-transcribed datasets may not be clipped, they have had their silences removed. Please be mindful of this while clipping.

Other Information

All data was taken from multiple publicly available YouTube sources. These datasets are not to be used for commercial purposes.