Whisper model fine-tuned for robust Assamese speech recognition

Researchers have developed a fine-tuned version of the Whisper model to improve Automatic Speech Recognition (ASR) for the Assamese language. The fine-tuned model, trained on the Mozilla Common Voice 24.0-Assamese corpus and optimized for resource-constrained environments using Tesla 4 GPUs, significantly outperformed the zero-shot baseline. The new system achieved substantial reductions in Word Error Rate (WER), Character Error Rate (CER), Match Error Rate (MER), and Word Infomation Loss (WIL), alongside notable improvements in BLEU and METEOR scores for semantic evaluation. AI

IMPACT Improves accessibility of speech technology for low-resource languages, potentially enabling new applications.

RANK_REASON The cluster contains an academic paper detailing a new method for improving speech recognition for a low-resource language using a fine-tuned existing model. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Assamese
• BLEU
• Character Error Rate
• Match Error Rate
• METEOR
• Mozilla Common Voice 24.0-Assamese
• Tesla 4 Graphics Processing Units
• Whisper
• word error rate
• Word Infomation Loss

• paper
• model release

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