Navigation

Your account

OPPORTUNITY DETAILS

Mozilla Lost in Transcription: Advancing Speech Recognition for Underserved Linguistic Contexts

Transcribe code-switched conversations across three language tracks. $20,000 total prize pool; submissions close October 2 at 23:59 UTC.

Algorithm competitionClosed · ArchivedMozilla Data Collective / Mozilla Foundation; operated by DrivenData · Online
On this pageEligibility and key factsAll deadlinesAbout this opportunityRules and deliverablesSources

Eligibility and key facts

Eligibility to be confirmed The summary below does not confirm your eligibility. Check eligible regions, age, language, and other requirements.

Who may be interested
Developers, Data scientists, Researchers
Team size
Individuals or teams of natural persons; one account and one team per person. Entities cannot enter; team-size cap not verified.
Location
Online; subject to official territorial and legal restrictions
Eligible regions
To be confirmed; online does not imply worldwide eligibility
Entry fee
Free
Prizes and support
$20,000 total: $4,000 first and $2,000 second in each of three tracks, plus $2,000 for the best average WER across all tracks with identical team membership.

All deadlines

Submission deadlineUTCClosed

Original date: All three official track pages: Competition End Date: Oct. 2, 2026, 11:59 p.m. UTC.

This deadline has passed and cannot be added.

Already registered? Check your submission deadline. Saving an event does not register you.

About this opportunity

An online speech-recognition challenge series from Mozilla Data Collective, operated by DrivenData. Separate tracks cover Indonesian-Javanese, Spanish-Nahuatl and North American Spanish-English conversations, evaluated with word error rate (WER). Participants may enter individual tracks; the cross-track bonus requires identical team membership across all three. All three pages displayed Join competition when checked. No separate early entry cutoff was found; entrants must still register and accept official rules.

Speech recognitionNatural language processingMultilingual AI

Rules and deliverables

Participation rules

  • Entrants must be at least 18 and of local legal majority. Entities/organizations cannot enter. OFAC and local-law restrictions apply; Russia-based entrants/team members and affiliated persons/family cannot win prizes.
  • Models must run locally. Do not send competition data to third-party model APIs. Authorized external training data must be shared through MDC after the competition as required by the rules.
  • Winning code must use MPL 2.0 and include reproduction documentation.
  • No separate early registration deadline was found; the submission deadline is not asserted to be a distinct entry cutoff.

What to prepare

  • Submit submission.zip to the selected track, with main.py and model weights at its root.
  • Run inference within two hours in the official offline Python 3.12 container and generate submission.csv with audio_filename and transcript columns.

Sources

Source platform
Mozilla Data Collective Competitions / DrivenData
Source domain
competitions.mozilladatacollective.com

The organizer’s published schedule and rules take precedence.