PhaseCoder: Microphone Geometry-Agnostic Spatial Audio Understanding for Multimodal LLMs

TL;DR

PhaseCoder is a transformer-based spatial audio encoder that is agnostic to microphone geometry, enabling multimodal LLMs to understand and reason about spatial audio from arbitrary microphone setups.

Contribution

It introduces a microphone geometry-agnostic spatial audio encoder that enhances multimodal LLMs with spatial reasoning capabilities.

Findings

Achieves state-of-the-art microphone-invariant localization performance.

Enables LLMs to perform complex spatial reasoning tasks.

Supports targeted transcription from arbitrary microphone arrays.

Abstract

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.