Lost In Transcription 2026: Mozilla speech recognition competition
Mozilla Data Collective pays a $20,000 prize pool for speech recognition models that handle bilingual speech. Entries close October 2, 2026.
Lost In Transcription is a speech recognition competition run by the Mozilla Data Collective. Its full name is Lost In Transcription: Advancing Speech Recognition for Underserved Linguistic Contexts. The prize pool is $20,000 and submissions close on October 2, 2026 at midnight UTC.
The competition has three tracks, one for each language pair: Spanish and English, Spanish and Nahuatl, and Indonesian and Javanese. Each track pays $4,000 for first place and $2,000 for second place. A bonus of $2,000 goes to the team with the lowest average word error rate across all three tracks. To qualify for the bonus, a team has to enter all three tracks with the same members.
The task is to build models that transcribe real bilingual speech, where speakers move between two languages inside one conversation. Teams may train on any other dataset in the Mozilla Data Collective catalogue.
Entrants must be at least eighteen and the age of majority where they live. They must be a legal resident of an eligible territory and must not be a target of United States trade sanctions. People in the Russian Federation cannot win prizes.
Winning solutions must be made available under the Mozilla Public License 2.0.