LLaMA-Omni2: LLM-based Real-time Spoken Chatbot with Autoregressive
  Streaming Speech Synthesis

Qingkai Fang; Yan Zhou; Shoutao Guo; Shaolei Zhang; Yang Feng

返回文献探索

Paper · arXiv 2505.02625

LLaMA-Omni2: LLM-based Real-time Spoken Chatbot with Autoregressive Streaming Speech Synthesis

Qingkai Fang, Yan Zhou, Shoutao Guo, Shaolei Zhang, Yang Feng

23 upvotesMay 5, 2025arXiv 预印本

AI 摘要

LLaMA-Omni 2, a series of speech language models with parameters ranging from 0.5B to 14B, achieves high-quality real-time speech interaction through a speech encoder and autoregressive streaming speech decoder, outperforming models like GLM-4-Voice with significantly less training data.

speech language modelsSpeechLMsQwen2.5speech encoderautoregressive streaming speech decoderspoken question answeringspeech instruction followingGLM-4-Voice

Abstract

Real-time, intelligent, and natural speech interaction is an essential part of the next-generation human-computer interaction. Recent advancements have showcased the potential of building intelligent spoken chatbots based on large language models (LLMs). In this paper, we introduce LLaMA-Omni 2, a series of speech language models (SpeechLMs) ranging from 0.5B to 14B parameters, capable of achieving high-quality real-time speech interaction. LLaMA-Omni 2 is built upon the Qwen2.5 series models, integrating a speech encoder and an autoregressive streaming speech decoder. Despite being trained on only 200K multi-turn speech dialogue samples, LLaMA-Omni 2 demonstrates strong performance on several spoken question answering and speech instruction following benchmarks, surpassing previous state-of-the-art SpeechLMs like GLM-4-Voice, which was trained on millions of hours of speech data.