Dhivehi Synthetic Speech Dataset with 2,627 Hours of Augmented Audio

1.26 million synthetic audio samples support research in Dhivehi, a low-resource language. The dataset was created by user alakxender and updated on October 15, 2025. It pairs Dhivehi sentences with waveforms generated through controlled synthesis, voice-cloning, and acoustic perturbations.

Use Cases

• Training automatic speech recognition models based on Dhivehi sentences paired with audio.
• Developing text-to-speech systems based on multi-speaker synthetic audio samples.
• Researching voice representation and speaker adaptation based on heavy acoustic perturbations.
• Improving model robustness in pronunciation and prosody based on controlled synthesis techniques.

Strengths

• Contains 1.26 million audio samples, providing a substantial volume of data.
• Offers approximately 2,627 hours of total audio duration.
• Focuses on Dhivehi, a language described as low-resource for AI/ML tasks.
• Includes samples from 82 speakers, suggesting multi-speaker diversity.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the synthetic generation methods described.

Provenance

Source

huggingface user alakxender

Collection Method

Generated through controlled synthesis, voice-cloning, and heavy acoustic perturbations.

Freshness

Last updated 2025-10-15 13:14:32; freshness should be verified.