LLaSO-Instruct
Multi-task instruction tuning dataset for the LLaSO framework, 13.5 million samples covering 20 speech tasks
Duration
None
Languages
1
Sample Rate
16 kHz
Published
2025-08
Description
1Instruction tuning component of the LLaSO open-source framework, containing 13.5 million multi-task instruction samples
2Covers 20 tasks distributed as: linguistic tasks 52%, semantic tasks 8%, paralinguistic tasks 40%
3Supports three interaction modes: text instruction + audio input, audio instruction + text input, audio-only
4Audio composition: 71% real-world audio, 29% synthesized speech
5Data sources include GigaSpeech, LibriSpeech, VoxCeleb1, Common Voice, MELD, CREMA-D, and other corpora
Language Details
| Language | Duration |
|---|---|
| English | None |
Publisher