Challenges / mcit
Khaliji Speech Recognition Challenge 2025
Transcribe conversational Gulf Arabic from government contact-centre calls, spanning Qatari, Bahraini, and Emirati accents with heavy English code-switching. Sadeem AI's winning system achieved 11.4 WER and was published to the marketplace as khaliji-asr, now powering voice services across three ministries.
Winning model published to the marketplace
Sadeem AI's Khaliji ASR passed the RAI pipeline and is live in the catalog.
Leaderboard
Scored on WER: lower is better| Rank | Team | Organisation | WER | Submissions |
|---|---|---|---|---|
| 1 | Sadeem AI | Sadeem AI | 11.4 | 38 |
| 2 | QCRI SpeechLab | HBKU | 12.1 | 45 |
| 3 | HBKU VoiceTech | HBKU | 13.8 | 31 |
| 4 | Wahaj Labs | Wahaj Labs | 15.2 | 19 |
| 5 | @abdulrahman_asr | independent | 16.7 | 23 |
| 6 | Falak Analytics | Falak Analytics | 17.9 | 12 |
| 7 | @leila_speech | independent | 19.3 | 15 |
| 8 | @vikram_audio | independent | 21.6 | 9 |
| 9 | HBKU MSc Cohort A | HBKU | 24.1 | 8 |
| 10 | @khalifa_dev | independent | 27.5 | 4 |
Note: this challenge is scored on WER (word error rate), so lower scores are better, Sadeem AI's winning 11.4 WER leads the board.
Submit an entry
Submission upload disabled
This challenge has concluded
Challenge closed
Teams must be registered on the platform. Submissions are evaluated on a held-out national test set and logged to the audit trail.