PTC-Bias improves speech LLM accuracy with phoneme-level bias correction
Researchers introduced PTC-Bias, a two-stage method for reducing errors in speech large language models (SpeechLLMs). The framework uses phoneme-level temporal competition to refine rare-word recognition. In the first stage, PTC Retrieval decodes phonemes frame-by-frame and narrows down candidate pronunciations into a shortlist. The second stage, PTC Correction, compares retrieved candidates…
Key points
- PTC-Bias uses phoneme-level temporal competition to filter and correct rare-word errors in SpeechLLMs
- Tests on LibriSpeech show up to 23.9% relative reduction in bias-word error rates for 2000-word lists
- Method requires no additional SpeechLLM forward passes, relying only on phoneme posteriors
Tests on the LibriSpeech dataset show PTC-Bias cuts bias-word error rates by up to 23.4% (test-clean) and 23.9% (test-other) compared to CTC-Filter, while keeping unweighted error rates stable. The method works with bias lists of up to 2000 words and two SpeechLLMs: Prompt-SLAM-ASR-7B and an unspecified second model. No extra SpeechLLM forward passes are needed, as both stages reuse phoneme posteriors.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Researchers release META, an episodic-memory trading agent for financial decisions · 1 src
- Epydemix Agent Framework automates epidemic modeling with AI agents · 1 src
- Researchers introduce tracer, a user simulator for AI behavior alignment · 1 src
- Researchers introduce Adversarial Closed-Loop training for Role-Playing AI agents · 1 src
- DEEPO improves hallucination in multimodal large language models · 1 src
Comments
via GitHub Discussions