Whisper-AuT: Domain-Adapted Audio Encoder for Efficient Audio-LLM Training

TL;DR

Whisper-AuT is a domain-adapted audio encoder, fine-tuned from Whisper-large-v3, that improves representation quality for music, environmental sounds, and speech, reducing training costs for audio-LLMs.

Contribution

The paper introduces Whisper-AuT, a fine-tuned version of Whisper-large-v3, optimized for diverse audio domains to enhance downstream audio-LLM performance.

Findings

+23.0% on ESC-50 environmental sound classification

+5.0% on GTZAN music genre classification

+0.7% on Speech Commands keyword spotting

Abstract

Audio-native large language models (audio-LLMs) commonly use Whisper as their audio encoder. However, Whisper was trained exclusively on speech data, producing weak representations for music and environmental sound. This forces downstream audio-LLMs to compensate through extensive training on large-scale non-speech data. We present Whisper-AuT, a domain-adapted audio encoder obtained by fine-tuning Whisper-large-v3 on a curated mixture of speech (80%), environmental sound (10%), and music (10%) totaling approximately 20M samples. The full encoder-decoder is trained end-to-end with a seq2seq captioning objective; the decoder is then discarded and only the encoder is retained. Linear probe evaluations show that Whisper-AuT achieves +23.0% on ESC-50 (environmental sound), +5.0% on GTZAN (music genre), and +0.7% on Speech Commands (keyword spotting) compared to the original Whisperlarge-v3 encoder. Whisper-AuT is designed as a drop-in replacement for Whisper in audio-LLM architectures, with the goal of reducing downstream training cost by providing stronger initial audio representations for non-speech domains.