arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2606.09366cs.CLeess.AS
收藏

文本就是一切?文本作为语音大语言模型的通用信息瓶颈

Is Text All You Need? Text as a Universal Information Bottleneck for Speech LLMs

The Chinese University of Hong Kong, Shenzhen(香港中文大学(深圳));Microsoft Corporation(微软公司);Microsoft Research Asia(微软亚洲研究院)

Ming-Hao Hsu, Yuxuan Hu, Shujie Liu, Jinyu Li, Yan Lu, Zhizheng Wu

AI总结:

提出Convex Gate(C-Gate)桥接语音与LLM,通过凸包约束将语音表示限制在LLM输入嵌入流形内,在ASR和情感识别上取得联合最优性能,并揭示几何结构而非离散性是关键设计因素。

摘要:

大型语言模型(LLM)为语音理解提供了强大的推理骨干,但将连续声学信号集成到冻结的LLM中仍然具有挑战性。现有的语音到LLM接口通常处于两个极端:要么强制近乎离散的令牌对齐,这有利于转录但丢失副语言信息;要么学习无约束的连续表示,这可能会偏离LLM的输入空间并降低自回归解码性能。在这项工作中,我们提出了Convex Gate(C-Gate),一种语音到LLM的桥接方法,通过架构凸包约束将所有语音表示限制在LLM的输入嵌入流形内。具体而言,每一帧被表示为令牌嵌入的凸组合,确保与预训练LLM的兼容性,同时保持连续表达能力。在自动语音识别(ASR)和情感识别任务中,C-Gate实现了强大的联合性能,在LibriSpeech上相对词错误率(WER)降低高达48.7%,同时匹配或超过单任务情感识别准确率。除了性能之外,我们的分析揭示了一个关键见解:信息不是由离散令牌身份携带,而是由嵌入空间中时间分辨的轨迹携带。因果干预证实,轨迹结构和与预训练嵌入流形的对齐对性能都至关重要。这些结果表明,几何结构而非令牌离散性是语音到LLM接口的基本设计因素,并为研究冻结LLM中的多模态集成提供了一个受控机制。我们发布了检查点、每个样本的输出、机制转储和干预套件以供复现。

英文摘要:

Large language models (LLMs) provide a powerful reasoning backbone for speech understanding, but integrating continuous acoustic signals into a pretrained LLM remains challenging. A central design question is how to tie speech representations to the LLM's text input space while preserving the continuous variation needed for different speech tasks. In this work, we propose Convex Gate (C-Gate), a speech-LLM bridge that represents each frame as a convex combination of frozen token embeddings. This architectural convex-hull constraint fixes the output codebook while allowing continuous mixtures learned from task supervision. With a Whisper-large-v3 encoder, a partially adapted Qwen2.5-7B-Instruct decoder, and joint training on speech recognition, emotion recognition, and spoken reasoning, C-Gate performs accurate transcription and emotion recognition, our representative semantic and acoustic tasks, through one shared decoder. On text-only benchmarks, the adapted decoder loses instruction-following and MMLU accuracy and matches or exceeds the original model on GSM8K and HellaSwag. We further design analyses and interventions on the trained model to examine where the bridge sequence carries task information and how the decoder reads it. These findings show how fixed text embeddings can support both semantic and acoustic readouts through continuous speech-conditioned mixtures.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑