otherWantedaudiospeechbilingualcode-switching
Seeking two bilingual speakers for a small recording set
What they need
If you and someone you know regularly switch between two languages, this could be a good fit. I need 30 short recordings: 15 from each of two consenting adult speakers, using the same language pair throughout. Any pair is welcome. Talk about ordinary things like errands, food, work or weekend plans, with a real language switch in every clip. Aim for 5-15 seconds each and include an exact transcript with timestamps showing where the language changes. Keep the hesitations and self-corrections in the transcript. The hard part is getting natural speech and checking the labels carefully; both speakers should be comfortable with both languages.
Requirements
formatzip
coverageOne consistent language pair across the dataset, 15 clips from each of 2 fluent adult speakers, and at least 5 everyday topics overall. Each clip must have at least one switch with meaningful speech in both languages. Contributor-authored prompts are allowed; repeated readings of the same sentence do not count as distinct examples.
acceptanceAll 30 clips must be intelligible, match their transcripts, satisfy the speaker balance and language-switch requirement, and contain no clipping that obscures words. Record in a reasonably quiet room without music or television.
deliverablesOne ZIP with 30 WAV or FLAC files, annotations.jsonl, README.txt, and a participant-consent attestation using anonymous speaker IDs. Each clip must be 5-15 seconds long. Use mono audio recorded at 16 kHz or higher; do not upsample a lower-quality source.
required fieldsclip_id, file_name, speaker_id, language_pair, verbatim_transcript, language_segments (start_seconds, end_seconds, language, text), topic, recording_date, duration_seconds, sample_rate_hz
annotation rulesWrite transcripts in the normal writing system of each language and document any transliteration convention. Include hesitations and corrections. Mark language-switch boundaries within approximately 0.25 seconds of the audible switch. Have a person fluent in both languages check every clip; this may be a contributing speaker.
consent and provenanceOriginal human recordings made for this bounty only; no text-to-speech, voice cloning, spliced speech, or reused corpora. Speakers must consent to redistribution and speech-recognition research/training. Use anonymous IDs, retain consent records, and include a dated attestation and reuse terms. Do not include personal contact information. AI transcription assistance is allowed only with full human verification.
$60.00
bounty reward per winner
0 submissions
Up to 1 winner
Posted by Shaurya Agarwal
Review target: 48 hours after a safe sample is ready.
35d 2h remaining