Arabic speech dataset improves recognition of child and L2 voices
AraYoungVoices: A Diverse L1/L2 Corpus of Arabic Child and Adolescent Speech
Computation and LanguageArtificial IntelligenceSound
Summary
Speech recognition systems usually work best for adult native speakers, but have trouble understanding children and people who speak Arabic as a second language. The paper presents AraYoungVoices, a large collection of recorded Arabic readings from children and teenagers, both native and non-native speakers. The authors tested existing speech recognition models on this dataset and found that recognizing second-language speakers, especially younger ones, is much harder. Training models specifically by age or combining all ages improves accuracy, and the models sometimes correct reading mistakes by aligning more with the expected prompts.
What this means in practice
- •For speech technology developers: Develop Arabic speech recognition systems that better understand children and second-language speakers across diverse dialects using the AraYoungVoices dataset.
- •For language learning app developers: Improve feedback and transcription accuracy in Arabic language learning apps for young and non-native speakers by incorporating models trained on AraYoungVoices data.
Authors
Shammur Absar Chowdhury, Zien Sheikh Ali, Houssam Eddine-Othman Lachemat, Hamdy Mubarak
Abstract
State-of-the-art ASR systems primarily target native adult speech, leading to substantial performance gaps for children, adolescents, and L2 speakers. We introduce AraYoungVoices, a 151.72-hour Arabic read-speech corpus from 286 speakers aged 7--18, comprising AraKids (7--12) and AraTeens (13--18). The corpus includes 146 native Arabic (L1) and 140 second-language (L2) speakers, with native speakers spanning Egyptian, Gulf, Levantine, and North African dialectal backgrounds and L2 speakers representing diverse linguistic backgrounds across the Americas, Asia, Africa, and Europe. We benchmark four pretrained ASR models under zero-shot and fine-tuned settings using unseen-speaker-$\&$-unseen-prompt (USUP) and unseen-speaker-$\&$-seen-prompt (USSP) evaluations. Results show that L2 speech remains substantially more challenging than L1 speech, with the largest errors observed mainly for younger L2 speakers. Age-specific fine-tuning improves the matched age group, while joint fine-tuning provides a stronger balance across populations. ASR hypotheses are also consistently closer to the standard reading prompt than to the verbatim transcription, particularly for L2 speech, suggesting partial normalization of reading deviations.