Authors: Liang-Hsuan Tseng, En-Pei Hu, Cheng-Han Chiang, Yuan Tseng, Hung-yi Lee, Lin-shan Lee, Shao-Hua Sun

Published on: February 06, 2024

Impact Score: 8.22

Arxiv code: Arxiv:2402.03988

Summary

• What is new: REBORN introduces a novel approach that combines reinforcement learning with iterative training to improve unsupervised automatic speech recognition.
• Why this is important: Learning the mapping between speech signals and text without paired data is challenging due to the variable length and unknown boundaries of word/phoneme segments.
• What the research proposes: REBORN, a method that alternates between training a segmentation model and a phoneme prediction model, using reinforcement learning for improving segmentation.
• Results: REBORN outperforms previous unsupervised ASR models on multiple datasets, including LibriSpeech, TIMIT, and non-English languages from Multilingual LibriSpeech.

Technical Details

Technological frameworks used: Iterative training with reinforcement learning for boundary segmentation.

Models used: Segmentation model and phoneme prediction model.

Data used: LibriSpeech, TIMIT, Multilingual LibriSpeech datasets.

Potential Impact

Speech recognition software providers, educational tech, and language learning platforms could benefit or be disrupted.

Want to implement this idea in a business?

We have generated a startup concept here: PhonemeGenie.