New Finnish Speech SSL Method Improves Accuracy Using ASR Confidence Filtering

Researchers have developed a new method for self-supervised learning (SSL) in Finnish speech by filtering raw broadcast audio based on Automatic Speech Recognition (ASR) confidence. This approach, applied to 2.1 million hours of unconstrained broadcast data, resulted in a 3-7% relative improvement in Word Error Rate (WER) compared to traditional language identification filtering, even with a smaller training set. The study also conducted a comparative analysis of Transformer, Macaron-Net, Conformer, and E-Branchformer architectures at matched parameter budgets, showing improvements over existing Finnish SSL baselines, particularly at the Base scale and for out-of-domain tasks. AI

IMPACT This research could lead to more efficient and accurate speech processing models, particularly for low-resource languages and noisy audio environments.

RANK_REASON The cluster contains an academic paper detailing a new method for speech self-supervised learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• Conformer
• E-Branchformer
• Finnish
• Hugging Face
• Macaron-Net
• transformer

• paper
• other

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