Hindi Tokens: Pre-Extracted Audio Codec Tokens for TTS Training

somu9's Hindi Tokens dataset contains 305,847 pre-extracted audio codec tokens for text-to-speech training. The data comprises 544.2 hours of Hindi audio, with an average sample duration of 6.4 seconds. It was last updated on June 2, 2026.

Use Cases

• Training Hindi text-to-speech models based on pre-extracted audio tokens.
• Fine-tuning audio codec tokenizers using a large collection of Hindi speech.
• Benchmarking TTS model performance on a dataset with 544.2 hours of audio.
• Researching audio token representations for speech synthesis in Indic languages.

Strengths

• Contains 305,847 individual audio samples.
• Offers a substantial 544.2 hours of Hindi audio data.
• Tokens are pre-extracted using the MOSS-Audio-Tokenizer-Nano codec at a 48,000 Hz sample rate.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

somu9/hindi dataset on Hugging Face.

Collection Method

Pre-extracted using the MOSS-Audio-Tokenizer-Nano codec.

Freshness

Last updated 2026-06-02 12:11:01; freshness should be verified.