SADA 2022
A speech recognition evaluation corpus for Saudi Arabian dialects, containing speech data in dialects such as Najdi and Hijazi.
Timeline (last 90 days)
On the SADA corpus, after replay-mix fine-tuning with roughly 90% Saudi dialect plus 7% English, WER on Najdi and Hijazi dialects dropped from 55.05% to 29.96%.
The SADA 2022 corpus marks unintelligible speech with the Arabic tag "غيرواضح" (unclear), and the model cannot output this tag, so every occurrence constitutes an unavoidable error.
Structural checks retained 103,559 of 125,490 utterances, i.e. 133.7 hours, 82.5% of the starting set.
On the SADA corpus, the default UTMOS threshold of 3.0 would reject nearly all samples.
All Facts (4)
On the SADA corpus, after replay-mix fine-tuning with roughly 90% Saudi dialect plus 7% English, WER on Najdi and Hijazi dialects dropped from 55.05% to 29.96%.
50%UnverifiedThe SADA 2022 corpus marks unintelligible speech with the Arabic tag "غيرواضح" (unclear), and the model cannot output this tag, so every occurrence constitutes an unavoidable error.
50%UnverifiedStructural checks retained 103,559 of 125,490 utterances, i.e. 133.7 hours, 82.5% of the starting set.
50%UnverifiedOn the SADA corpus, the default UTMOS threshold of 3.0 would reject nearly all samples.
50%