Hindi Text-to-Speech Audio Dataset from YouTube Captions

somu9's Hindi TTS dataset contains 33,394 audio segments totaling 62.25 hours of speech. The data was collected from 149 YouTube videos using auto-generated closed captions for transcription. The audio is stored as 24kHz mono WAV files in sharded parquet format.

Use Cases

• Train text-to-speech models based on the Hindi audio segments and corresponding captions.
• Develop speech recognition systems based on the transcribed Hindi audio data.
• Benchmark audio processing pipelines based on the 24kHz mono WAV format.
• Analyze speech patterns and accents based on content sourced from YouTube videos.

Strengths

• Contains 33,394 distinct audio segments.
• Provides a substantial 62.25 hours of total audio duration.
• Audio is standardized at a 24kHz mono sample rate.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect geographic, dialectal, or content bias inherent to the source YouTube videos.

Provenance

Source

YouTube videos

Collection Method

Collected using auto-generated closed captions for transcription.

Freshness

Last updated 2026-05-31 18:06:06; freshness should be verified.