AINeutralarXiv – CS AI · 5h ago6/10
🧠
HybridCodec: Fast Dual-Stream, Semantically Enhanced Neural Audio Codec
HybridCodec presents a novel neural audio codec architecture that combines semantic and acoustic feature streams while distilling SSL representations, achieving 3x speedup over existing dual-stream models. The advancement addresses the growing demand for efficient audio tokenizers in multimodal large language models by improving semantic specialization and cross-lingual robustness.