반응형
[Paper 리뷰] StableToken: A Noise-Robust Semantic Speech Tokenizer for Resilient SpeechLLMs
StableToken: A Noise-Robust Semantic Speech Tokenizer for Resilient SpeechLLMsSemantic speech tokenizer는 meaning-irrelevant acoustic perturbation에 fragile 함StableTokenConsensus-driven mechanism을 통해 robustness를 향상특히 multi-branch architecture를 활용해 parallel processing을 지원논문 (ICLR 2026) : Paper Link1. IntroductionSpeech Large Language Model (SpeechLLM)은 discrete speech tokenizer를 통해 continuous audio..
Paper/Neural Codec
2026. 4. 28. 12:57
반응형