Model

LLaMA-Omni

An end-to-end speech-language model architecture that extends a pretrained language model into a speech interaction model through a speech encoder, speech adapter, and speech decoder. It eliminates the ASR-to-LLM-to-TTS cascade, reducing latency and preserving paralinguistic information.

Timeline (last 90 days)

Oct 28

The model uses LLaMA-Omni as its base architecture and adopts an end-to-end speech interaction design, avoiding the multi-stage ASR→LLM→TTS cascade of traditional cascaded approaches.

Unverified50%

All Facts (1)

Unverified

The model uses LLaMA-Omni as its base architecture and adopts an end-to-end speech interaction design, avoiding the multi-stage ASR→LLM→TTS cascade of traditional cascaded approaches.

50%

Source Articles

The First Open-Source Speech Language Model for Basque: Breaking the Low-Resource Bottleneck with Synthetic DataOct 28