AudioLM: a Language Modeling Approach to Audio Generation

TL;DR

AudioLM introduces a novel audio generation framework that models audio as a language, enabling high-quality, coherent, and long-term audio synthesis across speech and music without transcripts.

Contribution

It presents a hybrid tokenization scheme and a language modeling approach that improves long-term coherence and quality in audio generation tasks.

Findings

Generates natural, coherent speech continuations maintaining speaker identity.

Produces high-quality piano music continuations without symbolic data.

Outperforms existing methods in long-term audio consistency.

Abstract

We introduce AudioLM, a framework for high-quality audio generation with long-term consistency. AudioLM maps the input audio to a sequence of discrete tokens and casts audio generation as a language modeling task in this representation space. We show how existing audio tokenizers provide different trade-offs between reconstruction quality and long-term structure, and we propose a hybrid tokenization scheme to achieve both objectives. Namely, we leverage the discretized activations of a masked language model pre-trained on audio to capture long-term structure and the discrete codes produced by a neural audio codec to achieve high-quality synthesis. By training on large corpora of raw audio waveforms, AudioLM learns to generate natural and coherent continuations given short prompts. When trained on speech, and without any transcript or annotation, AudioLM generates syntactically and semantically plausible speech continuations while also maintaining speaker identity and prosody for unseen speakers. Furthermore, we demonstrate how our approach extends beyond speech by generating coherent piano music continuations, despite being trained without any symbolic representation of music.