Tag: Cross-Modal Mapping

All the papers with the tag "Cross-Modal Mapping".

• VITA-Audio: Fast Interleaved Cross-Modal Token Generation for Efficient Large Speech-Language Model

  grok-3-latest

  Score: 0.57

  Published:2025年5月6日 at  at 17:59

  #LLM, #Cross-Modal Mapping, #Speech Generation, #Token Prediction, #Real-Time Interaction

  VITA-Audio通过轻量级MCTP模块和四阶段训练策略，在端到端语音模型中实现首次音频token零延迟生成，显著提升推理速度并在ASR、TTS、SQA任务上达到开源模型最优性能，为实时语音交互设定了新标准。