Dolly-Audio: 1,000 Hours of Multi-Speaker Vietnamese Speech

Dolly-Audio contains 1,000 hours of professionally cleaned Vietnamese speech audio featuring 152 speakers from various regions. Created by the Dolly AI Team and updated in December 2024, the corpus is designed to support speech synthesis and recognition research. It includes both audio recordings and corresponding text transcripts across multiple Vietnamese dialects.

Use Cases

• Multi-speaker Text-to-Speech (TTS) synthesis using the audio modality and Vietnamese text transcripts
• Automatic Speech Recognition (ASR) training across 152 unique speaker voices
• Regional dialect classification using audio samples from different Vietnamese regions

Strengths

• 1,000 hours of audio recordings
• 152 unique speakers
• Coverage of multiple Vietnamese regional dialects
• Professionally cleaned audio quality

Limitations

• Metadata tags indicate the presence of synthetic data which may affect naturalness in some applications
• Lack of detailed column-level documentation for speaker metadata

Provenance

Source

Dolly AI Team (vuhoanhuy)

Collection Method

synthetic

Freshness

Last updated December 2024.

Geography

Vietnam