IISc Mono Hindi Female: 54-Hour Studio-Quality Speech Dataset

54 hours 54 minutes of studio-quality Hindi speech from a single professional female voice artist, recorded at 48kHz and 24-bit. The dataset contains 22,058 utterances, split into 21,662 for training and 396 for evaluation, and was created by the Indian Institute of Science (IISc) SYSPIN project. It was uploaded to Hugging Face by user 'somu9' and last updated on April 15, 2026.

Use Cases

• Train text-to-speech models based on the 54 hours of high-fidelity Hindi audio.
• Fine-tune voice synthesis systems based on the single professional female speaker profile.
• Benchmark speech synthesis quality based on the dedicated 396-utterance evaluation set.
• Develop monolingual Hindi speech applications based on the studio-quality recordings.

Strengths

• Large total duration of 54 hours 54 minutes 44 seconds of audio.
• High-quality audio specifications: 48kHz, 24-bit, mono recordings.
• Utterances from a single professional speaker with 21 years of experience.
• Pre-defined split with 21,662 training and 396 evaluation utterances.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Dataset is monolingual and single-speaker, limiting diversity for some applications.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Indian Institute of Science (IISc) SYSPIN Project

Collection Method

Studio recordings by a professional voice artist.

Freshness

Last updated 2026-04-15 15:10:58; freshness should be verified.

Geography

Likely India, given the language and source institution.