Soundwave: Less is More for Speech-Text Alignment in LLMs

2025/02/18 by Zhang, Yuhao, Liu, Zhiheng, Bu, Fan +3 · 5 citations
#Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #Sound (cs.SD)

paper · doi:10.48550/arxiv.2502.12900

Abstract

Existing end-to-end speech large language models (LLMs) usually rely on large-scale annotated data for training, while data-efficient training has not been discussed in depth. We focus on two fundamental problems between speech and text: the representation space gap and sequence length inconsistency. We propose Soundwave, which utilizes an efficient training strategy and a novel architecture to address these issues. Results show that Soundwave outperforms the advanced Qwen2-Audio in speech translation and AIR-Bench speech tasks, using only one-fiftieth of the training data. Further analysis shows that Soundwave still retains its intelligence during conversation. The project is available at https://github.com/FreedomIntelligence/Soundwave.

Cited by

• HPSU: A Benchmark for Human-Level Perception in Real-World Spoken Speech Understanding
• Towards General Auditory Intelligence: Large Multimodal Models for Machine Listening and Speaking
• M-CIF: Multi-Scale Alignment For CIF-Based Non-Autoregressive ASR
• EchoX: Towards Mitigating Acoustic-Semantic Gap via Echo Training for Speech-to-Speech LLMs
• SageLM: A Multi-aspect and Explainable Large Language Model for Speech Judgement

Related

• Closing the Gap Between Text and Speech Understanding in LLMs
• TASU: Text-Only Alignment for Speech Understanding
• TESU-LLM: Training Speech-LLMs Without Speech via Unified Encoder Alignment
• DeSTA2.5-Audio: Toward General-Purpose Large Audio Language Model with Self-Generated Cross-Modal Alignment
• From Alignment to Advancement: Bootstrapping Audio-Language Alignment with Synthetic Data