viVoice: Enabling Vietnamese Multi-Speaker Speech Synthesis

For a comprehensive description, please visit https://github.com/thinhlpg/viVoice This dataset is licensed under CC-BY-NC-SA-4.0 and is intended for research purposes only.

Key Features and Statistic 📊

• All audio is cleaned from noise and music.
• Clean cuts are made at the beginning and end of sentences to eliminate any unnecessary silences or disruptions, while avoiding cutting in the middle of words.
• Sourced from 186 YouTube channels, with channel IDs included for transparency.
• Number of samples: 887,772
• Total duration: 1,016.97 hours
• Sampling rate: 24 kHz
• Number of splits: 1 (train only)
• Size: 169 GBs
• Gender distribution of speakers: 61.3% ± 3.02% male (manually estimated from a sample of 1,000 with a 95% confidence interval)
• Estimated transcription error rate: 1.8% ± 0.82% (manually estimated from a sample of 1,000 with a 95% confidence interval)
  • This metric is for quick reference purposes only; users of this dataset should carefully inspect it to ensure it meets your requirements.
  • The error rate only accounts for sentences with mistranscriptions (more or fewer words than expected).
  • Other errors, such as missing punctuation or incorrect but phonetically similar words, are not counted.