Hindi Speech Instruct: 10 Multi-Turn Hindi Conversations for Speech Language Models

Hindi Speech Instruct is a multi-turn Hindi conversational dataset for training speech language models, created by author somu9. It contains 10 conversations with a total of 25 user audio turns paired with 25 assistant text responses. The dataset was last updated on 2026-06-17.

Use Cases

• Train speech-to-text models based on Hindi user audio turns.
• Develop conversational AI assistants based on paired user audio and assistant text responses.
• Fine-tune language models for Hindi dialogue generation based on the multi-turn conversation structure.
• Study multi-modal interaction patterns in Hindi based on the audio-text pairing.

Strengths

• Contains 10 complete multi-turn conversations, providing context for dialogue modeling.
• Pairs 25 user audio files with 25 assistant text responses, enabling audio-text alignment tasks.
• Provides conversation statistics including average user text length (32.4 chars) and assistant text length (94.0 chars).

Limitations

• Dataset scale is small, with only 10 total conversations and 25 user audio turns.
• 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 on Hugging Face

Collection Method

Likely collected or synthesized for training speech language models.

Freshness

Last updated 2026-06-17 18:51:54; freshness should be verified.