OpenPA: Forced-alignment-free Multi-granular Pronunciation Assessment with Mixed-transcript Adaptation
OpenPA uses Whisper cross-attention to score pronunciation at word and utterance level without forced alignment.
OpenPA is an automatic pronunciation assessment framework that uses Whisper cross-attention to predict word- and utterance-level scores without forced alignment. Mixed-transcript adaptation and adaptive substitution weighting let one model handle closed- and open-response transcripts. Trained only on speechocean762, it achieved the strongest overall results among systems tested on in-domain closed and simulated open-response conditions, and held performance zero-shot on MultiPA. A cross-task evaluation on oral-cancer speech suggests use beyond non-native speech.
- Joint word and utterance scores without forced alignment
- Trained only on the speechocean762 corpus
- Led evaluated systems on closed and open-response tests
- Zero-shot scores held on the MultiPA benchmark
From the original article
1 min read at arxiv.orgRecent advances in automatic pronunciation assessment (APA) have driven a shift from scripted to free-form assessment. However, existing methods often rely on multi-stage pipelines with forced alignment and separate feature extractors, limiting their practicality across response settings.
By Fu-An Chao · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.