MoWE-Audio: Multitask AudioLLMs with Mixture of Weak Encoders

TL;DR

This paper introduces MoWE-Audio, a multitask AudioLLM framework that uses a mixture of lightweight encoders to improve feature extraction and performance across diverse audio tasks.

Contribution

It proposes integrating mixtures of weak encoders into AudioLLMs, enhancing their capacity to handle new tasks without significantly increasing model size.

Findings

MoWE-Audio improves multi-task performance.

The mixture of encoders broadens task applicability.

Enhanced feature extraction with lightweight encoders.

Abstract

The rapid advancements in large language models (LLMs) have significantly enhanced natural language processing capabilities, facilitating the development of AudioLLMs that process and understand speech and audio inputs alongside text. Existing AudioLLMs typically combine a pre-trained audio encoder with a pre-trained LLM, which are subsequently finetuned on specific audio tasks. However, the pre-trained audio encoder has constrained capacity to capture features for new tasks and datasets. To address this, we propose to incorporate mixtures of `weak' encoders (MoWE) into the AudioLLM framework. MoWE supplements a base encoder with a pool of relatively light weight encoders, selectively activated based on the audio input to enhance feature extraction without significantly increasing model size. Our empirical results demonstrate that MoWE effectively improves multi-task performance, broadening the applicability of AudioLLMs to more diverse audio tasks.