Skip to content

Fact · Languages and accents

The 2023 Interspeech paper found that diacritizing inside the speech model outperformed diacritizing the text afterward, especially when the model was fine-tuned on manually diacritized transcripts.

Checked against its source on October 17, 2026

In the source's own words

“ASR diacritization significantly outperforms text-based diacritization, particularly when the ASR model is fine-tuned with manually diacritized transcripts”

Diacritic Recognition Performance in Arabic ASR, Hanan Aldarmaki and Ahmad Ghannam, Interspeech 2023, 2023, Abstract.

Open the original
What it measures
The authors compared three training conditions: manual diacritics, automatic diacritics and no diacritics, and measured diacritic recognition apart from overall word accuracy.
What it does not show
A research comparison on the authors' own training data. It shows the direction of the effect, not how a consumer app would perform.

Source

Cited on

Adding diacritics inside the model beat adding them to text afterward | Yap