Falcon ASR beats top speech models with 20.92% Arabic WER
tii releases falcon-asr, a 1.6b parameter arabic speech recognition model with 20.92% wer, outperforming competitors. this impacts arabic asr applications.
the technology innovation institute (tii) introduced falcon-asr, a 1.6 billion parameter speech recognition model focusing on arabic, particularly the emirati dialect. it achieved a 20.92% average word error rate (wer) across six arabic test sets, surpassing the best published result of 23.17%. in internal emirati evaluations, falcon-asr recorded a 22.73% wer and 10.19% character error rate (cer), outperforming competitors like audar-asr-v1-turbo and qwen3-omni-30b-a3b-instruct. the model also supports english, french, spanish, and portuguese, and achieved a 5.74% mean wer on seven public english test sets.