OpenPA: Forced-alignment-free Multi-granular Pronunciation Assessment with Mixed-transcript Adaptation
arXiv cs.CL · 10 Oct 2026, 02:35 AEST
OpenPA is a new framework for automatic pronunciation assessment that eliminates the need for forced alignment. It achieves strong performance across various response settings by using a single model for both word- and utterance-level scoring.
Key facts
- Recent advances in automatic pronunciation assessment (APA) have driven a shift from scripted to free-form assessment.
- We propose OpenPA, a novel APA framework that leverages Whisper's cross-attention mechanism to jointly predict word- and utterance-level pronunciation scores without forced alignment.
- Trained solely on speechocean762, OpenPA achieves the strongest overall performance among evaluated systems under in-domain closed and simulated open-response conditions.
Read aloud
OpenPA is a new framework for automatic pronunciation assessment that eliminates the need for forced alignment. It achieves strong performance across various response settings by using a single model for both word- and utterance-level scoring. Recent advances in automatic pronunciation assessment APA have driven a shift from scripted to free-form assessment. We propose OpenPA, a novel APA framework that leverages Whisper's cross-attention mechanism to jointly predict word- and utterance-level pronunciation scores without forced alignment. Trained solely on speechocean762, OpenPA achieves the strongest overall performance among evaluated systems under in-domain closed and simulated open-response conditions.
Sources in this cluster
- OpenPA: Forced-alignment-free Multi-granular Pronunciation Assessment with Mixed-transcript Adaptation
arXiv cs.CL · 10 Oct 2026, 02:35 AEST