Saved in:
Bibliographic Details
Main Authors: Dementyev, Artem, Zulfikar, Wazeer, Hersek, Sinan, Getreuer, Pascal, Kumar, Anurag, Kumar, Vivek
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.21124
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917230628831232
author Dementyev, Artem
Zulfikar, Wazeer
Hersek, Sinan
Getreuer, Pascal
Kumar, Anurag
Kumar, Vivek
author_facet Dementyev, Artem
Zulfikar, Wazeer
Hersek, Sinan
Getreuer, Pascal
Kumar, Anurag
Kumar, Vivek
contents Current multimodal LLMs process audio as a mono stream, ignoring the rich spatial information essential for embodied AI. Existing spatial audio models, conversely, are constrained to fixed microphone geometries, preventing deployment across diverse devices. We present PhaseCoder, a transformer-only spatial audio encoder that is agnostic to microphone geometry. PhaseCoder takes raw multichannel audio and microphone coordinates as inputs to perform localization and produces robust spatial embeddings. We demonstrate that Gemma 3n LLM can be fine-tuned to reason over "Spatial Audio Tokens" produced by PhaseCoder. We show our encoder achieves state-of-the-art results on microphone-invariant localization benchmarks and, for the first time, enables an LLM to perform complex spatial reasoning and targeted transcription tasks from an arbitrary microphone array.
format Preprint
id arxiv_https___arxiv_org_abs_2601_21124
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle PhaseCoder: Microphone Geometry-Agnostic Spatial Audio Understanding for Multimodal LLMs
Dementyev, Artem
Zulfikar, Wazeer
Hersek, Sinan
Getreuer, Pascal
Kumar, Anurag
Kumar, Vivek
Sound
Artificial Intelligence
Audio and Speech Processing
Current multimodal LLMs process audio as a mono stream, ignoring the rich spatial information essential for embodied AI. Existing spatial audio models, conversely, are constrained to fixed microphone geometries, preventing deployment across diverse devices. We present PhaseCoder, a transformer-only spatial audio encoder that is agnostic to microphone geometry. PhaseCoder takes raw multichannel audio and microphone coordinates as inputs to perform localization and produces robust spatial embeddings. We demonstrate that Gemma 3n LLM can be fine-tuned to reason over "Spatial Audio Tokens" produced by PhaseCoder. We show our encoder achieves state-of-the-art results on microphone-invariant localization benchmarks and, for the first time, enables an LLM to perform complex spatial reasoning and targeted transcription tasks from an arbitrary microphone array.
title PhaseCoder: Microphone Geometry-Agnostic Spatial Audio Understanding for Multimodal LLMs
topic Sound
Artificial Intelligence
Audio and Speech Processing
url https://arxiv.org/abs/2601.21124