IRIS-Data is a dataset for training instruction-aware speech retrieval models. Each example contains an instruction, a spoken query, a positive document, negative documents, and metadata for the retrieval criteria. The data comes from 14 sources listed below.

Files

File	Contents
`train.jsonl`	890,278 training examples
`val.jsonl`	9,653 validation examples
`audio/audio-*.tar`	391,468 WAV recordings in 34 archives

The WAV files are not resampled or re-encoded for this release.

Download

After access is approved, download with your authenticated Hugging Face account:

hf download lca0503/IRIS-Data --repo-type dataset --local-dir IRISData

for archive in IRISData/audio/audio-*.tar; do
    tar -xf "$archive" -C IRISData
done

Extraction creates IRISData/audio/<id>.wav. Audio paths in both JSONL files are relative to IRISData.

Example format

Field	Description
`instruction`	Retrieval instruction
`spoken_query`	Relative path to the query audio
`positive_document`	Relative path to the positive audio
`negative_documents`	List of relative paths to negative audio
`select_criteria`	Criteria for selecting positive and negative documents
`spoken_query_metadata`	Query metadata
`positive_document_metadata`	Positive document metadata
`negative_documents_metadata`	Metadata for each negative document
`source`	Source dataset identifier

Metadata includes speaker, gender, emotion, accent, sentiment, environmental_sound, dialogue_id, topic_id, and text. Missing attributes are null.

Sources and terms

Each source keeps its license and permission requirements. Access to IRIS-Data does not replace those requirements. See the source pages below for terms and attribution.

Source identifier	Source
`crema_d`	CREMA-D
`emns`	EMNS
`emov_db`	EmoV-DB
`esd`	Emotional Speech Dataset
`l2arctic`	L2-ARCTIC
`librispeech_c`	LibriSpeech-C recipe, LibriSpeech, and MS-SNSD
`meld`	MELD
`msp_improve`	MSP-IMPROV
`multidialog`	MultiDialog
`spokenwoz`	SpokenWOZ
`synthetic`	MS MARCO triplets from sentence-transformers/msmarco-bm25, synthesized with Gemini 2.5 Pro TTS, Gemini 3.1 Flash TTS, and Qwen3 TTS; background sounds from UrbanSound8K
`urbansound8k`	UrbanSound8K
`vocalsound`	VocalSound
`voxceleb1`	VoxCeleb1

Downloads last month: 240

Total file size:

167 GB