IntrinsicVoice: Empowering LLMs with Intrinsic Real-time Voice Interaction Abilities

TL;DR

IntrinsicVoice tackles the latency and modality-gap challenges of voice-enabled LLMs by introducing GroupFormer, which compresses speech token sequences to near-text lengths and enables non-autoregressive group-level generation. A cross-modal training strategy and a large speech-to-speech dataset (Intrinsic_500k) deepen semantic alignment between speech and text. Experiments show real-time (<100 ms) speech responses with competitive speech quality and improved content fidelity, outperforming SpeechGPT baselines in multi-turn scenarios. The work demonstrates a practical pathway to intrinsic, low-latency voice interaction for LLMs and provides resources and architectures to scale this capability.

Abstract

Current methods of building LLMs with voice interaction capabilities rely heavily on explicit text autoregressive generation before or during speech response generation to maintain content quality, which unfortunately brings computational overhead and increases latency in multi-turn interactions. To address this, we introduce IntrinsicVoic,e an LLM designed with intrinsic real-time voice interaction capabilities. IntrinsicVoice aims to facilitate the transfer of textual capabilities of pre-trained LLMs to the speech modality by mitigating the modality gap between text and speech. Our novelty architecture, GroupFormer, can reduce speech sequences to lengths comparable to text sequences while generating high-quality audio, significantly reducing the length difference between speech and text, speeding up inference, and alleviating long-text modeling issues. Additionally, we construct a multi-turn speech-to-speech dialogue dataset named \method-500k which includes nearly 500k turns of speech-to-speech dialogues, and a cross-modality training strategy to enhance the semantic alignment between speech and text. Experimental results demonstrate that IntrinsicVoice can generate high-quality speech response with latency lower than 100ms in multi-turn dialogue scenarios. Demos are available at https://instrinsicvoice.github.io/.

Figures (3)

• Figure 1: Left: The overview of the IntrinsicVoice model architecture. Right: Illustration of the proposed GroupFormer model structure.
• Figure 2: Prompt template of IntrinsicVoice
• Figure 3: Case of multi-turn QA result