PAL: Probing Audio Encoders via LLMs -- Audio Information Transfer into LLMs

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Alex, Tony, Suharitdamrong, Wish, Atito, Sara, Mustafa, Armin, Jackson, Philip J. B., Razzak, Imran, Awais, Muhammad
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910008085577728
author Alex, Tony
Suharitdamrong, Wish
Atito, Sara
Mustafa, Armin
Jackson, Philip J. B.
Razzak, Imran
Awais, Muhammad
author_facet Alex, Tony
Suharitdamrong, Wish
Atito, Sara
Mustafa, Armin
Jackson, Philip J. B.
Razzak, Imran
Awais, Muhammad
contents Integration of audio perception into large language models (LLMs) is an emerging research area for enabling machine listening applications, yet efficient transfer of rich audio semantics from audio encoders to LLMs remains underexplored. The most widely used integration paradigm projects audio-encoder output tokens into the LLM input space (e.g., via an MLP or a Q-Former) and then prepends or inserts them into the text token sequence. We refer to this generic scheme as Prepend to the LLM's input token space (PLITS) integration. We propose an efficient alternative, Lightweight Audio LLM Integration (LAL). LAL injects audio representations solely through the attention mechanism at selected LLM layers, bypassing the feed-forward module. It encodes rich audio semantics at an appropriate level of abstraction for integration into different transformer blocks, substantially reducing computational overhead compared to existing approaches. We further introduce PAL, a hybrid integration approach for efficiently Probing Audio encoders via LLM. PAL applies PLITS only to a compact set of summary tokens while integrating the full audio token sequence via LAL. Under an identical training curriculum, LAL consistently matches or outperforms existing integration approaches across multiple base LLMs and tasks, with improvements of up to 30% over a strong PLITS baseline, while reducing memory usage by about 60% and increasing throughput by about 190%. Moreover, PAL matches or exceeds PLITS performance while offering substantially better computational and memory efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2506_10423
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PAL: Probing Audio Encoders via LLMs -- Audio Information Transfer into LLMs
Alex, Tony
Suharitdamrong, Wish
Atito, Sara
Mustafa, Armin
Jackson, Philip J. B.
Razzak, Imran
Awais, Muhammad
Sound
Artificial Intelligence
Computation and Language
Audio and Speech Processing
Integration of audio perception into large language models (LLMs) is an emerging research area for enabling machine listening applications, yet efficient transfer of rich audio semantics from audio encoders to LLMs remains underexplored. The most widely used integration paradigm projects audio-encoder output tokens into the LLM input space (e.g., via an MLP or a Q-Former) and then prepends or inserts them into the text token sequence. We refer to this generic scheme as Prepend to the LLM's input token space (PLITS) integration. We propose an efficient alternative, Lightweight Audio LLM Integration (LAL). LAL injects audio representations solely through the attention mechanism at selected LLM layers, bypassing the feed-forward module. It encodes rich audio semantics at an appropriate level of abstraction for integration into different transformer blocks, substantially reducing computational overhead compared to existing approaches. We further introduce PAL, a hybrid integration approach for efficiently Probing Audio encoders via LLM. PAL applies PLITS only to a compact set of summary tokens while integrating the full audio token sequence via LAL. Under an identical training curriculum, LAL consistently matches or outperforms existing integration approaches across multiple base LLMs and tasks, with improvements of up to 30% over a strong PLITS baseline, while reducing memory usage by about 60% and increasing throughput by about 190%. Moreover, PAL matches or exceeds PLITS performance while offering substantially better computational and memory efficiency.
title PAL: Probing Audio Encoders via LLMs -- Audio Information Transfer into LLMs
topic Sound
Artificial Intelligence
Computation and Language
Audio and Speech Processing
url https://arxiv.org/abs/2506.10423