Hindi Anushka: Hindi Text-to-Speech Audio from YouTube, 600 Hours

A Hindi text-to-speech dataset containing 332,247 audio segments totaling 599.91 hours. The audio was collected from 378 YouTube videos using auto-generated closed captions for transcription. The dataset was created by author somu9 and last updated on 2026-05-29.

Use Cases

• Train Hindi text-to-speech models based on the audio and transcription data.
• Fine-tune speech synthesis models using the 24kHz mono audio format.
• Benchmark automatic speech recognition systems on Hindi audio with auto-generated captions.
• Develop voice cloning applications for Hindi using the segmented audio samples.

Strengths

• Large scale with 332,247 audio segments.
• Substantial total duration of 599.91 hours.
• Audio is provided in a consistent 24kHz mono format.
• Segments have an average duration of 6.5 seconds, which is suitable for many ML tasks.

Limitations

• Transcriptions rely on YouTube's auto-generated closed captions, which may contain errors.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset's source is limited to 378 YouTube videos, which may introduce content bias.

Provenance

Source

YouTube videos.

Collection Method

Collected from YouTube using auto-generated closed captions for transcription.

Freshness

Last updated 2026-05-29 10:57:32; freshness should be verified.