You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Luel Conversational Speech Samples

Multi-speaker conversational audio for ASR / TTS / dialogue research, organized into three quality tiers.

Tiers

Tier Folder What it is
Standard standard/ Curated, QA-passed conversational audio in 9 well-resourced languages. The canonical quality bar.
Noisy noisy/ Conversational audio recorded in non-ideal acoustic environments (crowd, traffic, animals, etc.). Useful for training noise-robust ASR. Currently English-only.
Low-resource low_resource/ Pilot bucket for low-resource languages. Per-sample quality varies more than the standard tier. Most sessions are audio-only; a small subset has transcripts.

Each tier has its own README inside the folder with details.

Per-session file convention (all tiers)

<tier>/<lang>/<session_id>/
    <session_id>.wav            # equal-weight host+guest mono mix
    <session_id>_speaker1.wav   # host track (isolated)
    <session_id>_speaker2.wav   # guest track (isolated)
    <session_id>.json           # metadata + (optional) speakerTranscript

Speaker tracks were captured simultaneously on independent channels — diarization is built into the data, not inferred. Both speaker tracks share an identical timeline and duration.

JSON schema

  • name, audioURL, speaker1_audioURL, speaker2_audioURL
  • durationSeconds, language, category, numSpeakers, sampleRate
  • metadata.* — language, country, region, environment, device, topic
  • transcript (string, joined from speakerTranscript)
  • speakerTranscript — array of {speaker, text, startTime, endTime, words[]} (when transcripts are available)

Filler words, stutters, false starts, and repeated words are preserved verbatim in transcripts.

Languages by tier

Standard: English (American + Global), Hindi, Punjabi, Marwari, Urdu, German, French, Japanese, Spanish.

Noisy: English (American).

Low-resource: Afrikaans, Akan/Twi, Amharic, Cebuano, Hausa, Hiligaynon, Igbo, Ilocano, Sepedi, Pashto, Shona, Sesotho, Tagalog, Tswana, Xhosa, Yoruba, Zulu.

Citation / contact

See LICENSE for usage terms. For questions or licensing, reach out via the Luel website.

Downloads last month
81

Collection including Luel-ai/luel-conversational-speech-samples