bharat-speech
Collection
Bharat-Speech is an open-source collection of speech datasets for building and evaluating ASR, TTS, and other speech AI systems for Indian languages. • 3 items • Updated
The dataset viewer is not available for this split.
External: Generic Unexpected error: Unexpected (temporary) at read, context: { url: https://huggingface.co/datasets/psk/vaani-asr/resolve/refs%2Fconvert%2Fparquet/bengali/train/0000.parquet, called: http_util::Client::send, service: huggingface, path: bengali/train/0000.parquet, range: 4-21128371 } => send http request, source: error sending request for url (https://huggingface.co/datasets/psk/vaani-asr/resolve/refs%2Fconvert%2Fparquet/bengali/train/0000.parquet)
Error code: UnexpectedError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Vaani ASR (transcribed subset)
Language-wise ASR dataset built from the quality-checked transcribed subset of the Vaani project (ARTPARK / IISc Bangalore) recordings: 172k+ short speech clips (16 kHz mono WAV) from 12+ Indian states with verbatim transcripts in native scripts.
- Each subset = primary spoken language (first language listed by annotators);
clips with under 500 examples per language are grouped in
other. - Columns:
audio,file_name,transcript(verbatim, may contain annotation tags such as<noise>,<horn>, and{word}code-switch markers),state,gender,languages(full spoken-language list, JSON). - Source data: Vaani project, IISc Bangalore & ARTPARK. Attribution: Vaani / ARTPARK-IISc. License follows the Vaani release (CC-BY-4.0).
- Downloads last month
- 925