Multilingual Indian Speech Corpus for Text-to-Speech

1,704 hours of speech data from 10,496 speakers across 22 Indian languages, derived from an automatic speech recognition dataset. Created by ai4bharat, this corpus was updated in March 2025 to support the development of text-to-speech models.

Use Cases

• Training text-to-speech models using the multilingual speech audio and corresponding transcriptions.
• Developing multi-speaker voice synthesis systems leveraging the 10,496 distinct speaker identities.
• Researching speaker adaptation and voice cloning across the 22 Indian language labels.
• Analyzing prosody and natural conversational speech patterns present in the audio recordings.
• Benchmarking speech synthesis quality for low-resource Indian languages using the provided high-quality audio.

Strengths

• Contains 1,704 hours of high-quality speech audio.
• Features a diverse speaker pool of 10,496 individuals.
• Covers a wide range of 22 Indian languages.

Limitations

• Derived from an ASR dataset, which may introduce biases from the original collection and transcription process.
• Specific details on speaker demographics, recording conditions, and data splits are not provided in the summary.

Provenance

Source

ai4bharat

Collection Method

Derived from an automatic speech recognition dataset.

Time Range

null

Freshness

Last updated March 2025.

Geography

India (covers 22 Indian languages)