New MoPE-LoRA framework improves accented English speech recognition

Researchers have developed a new parameter-efficient adaptation framework called Mixture of Phonetic Experts based Low-Rank Adaptation (MoPE-LoRA) for accented English speech recognition. Unlike previous methods that assign experts per accent, MoPE-LoRA organizes adaptation around phonetic categories, specifically manner-of-articulation classes. This approach allows the model to extend to unseen accents without requiring accent-specific modules. Experiments on the L2-ARCTIC dataset demonstrated that MoPE-LoRA significantly outperforms single LoRA and full fine-tuning, showing a 12.3% relative improvement in zero-shot evaluation. AI

IMPACT Offers a more scalable and transferable approach to accent adaptation in speech recognition systems.

RANK_REASON Academic paper detailing a new method for speech recognition. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• Conformer Models
• English speech recognition based on deep learning with multiple features
• L2-ARCTIC
• LoRA+
• MoPE-LoRA

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →