Introducing Falcon ASR
We’re introducing Falcon-ASR, our 1.6 billion parameter speech recognition model for Arabic, with a particular focus on the Emirati dialect.
Key points
- Arabic WER: 20.92% · Parameters: 1.6B · Emirati WER (TII evaluation): 22.73%
- In our evaluation, Falcon-ASR achieved an average word error rate of 20.92% across six Arabic test sets, compared with the best published result of 23.17% in the leaderboard snapshot we used.
- Arabic speech varies by region, speaker and setting.
- Our Falcon-ASR evaluation follows this protocol.
Sources (1)
- [1]Introducing Falcon ASRHugging Face Blog · Oct 7, 01:21 PM
We’re introducing Falcon-ASR, our 1.6 billion parameter speech recognition model for Arabic, with a particular focus on the Emirati dialect.
Arabic WER: 20.92% · Parameters: 1.6B · Emirati WER (TII evaluation): 22.73%
Extractive summary: sentences quoted from the sources.