analiza aiqbench
The paper tests whether language models trained on raw audio can compete with established lossless codecs at production bit depths (16 and 24-bit), not just the 8-bit setting studied before. Its main technical contribution, Trilobyte, is a byte-level tokenization scheme that makes 24-bit modeling tractable by changing vocabulary scaling from exponential to constant. Readers should note that the gains over FLAC are strongest at 8 and 16-bit and shrink at higher bit depths, so the practical advantage at full fidelity is more limited than the framing suggests.
Benchmarking Language Modeling for Lossless Compression of Full-Fidelity Audio
arXiv:2603.08683v3 Announce Type: replace-cross Abstract: Autoregressive "language" models (LMs) trained on raw waveforms can be repurposed for lossless audio compression, but prior work is limited to 8-bit audio, leaving open whether such approaches work for practical settings (16/24-bit) and can compete with existing codecs. We benchmark LM-based compression on full-fidelity audio across diverse domains (music, speech, bioacoustics), sampling rates (16kHz-48kHz), and bit depths (8, 16, 24-bit). Standard sample-level tokenization becomes intractable at higher bit depths due to vocabulary size (65K for 16-bit; 16.7M for 24-bit). We propose Trilobyte, a byte-level tokenization schema for full resolution audio, improving vocabulary scaling from $O(2^{b})$ to $O(1)$ and enabling the first tractable 24-bit LM-based lossless compression. While LMs consistently outperform FLAC and yield state-of-the-art compression at 8-bit and 16-bit, we observe that compression gains become more modest as bit depth increases beyond 8-bit.
Abstract:Autoregressive "language" models (LMs) trained on raw waveforms can be repurposed for lossless audio compression, but prior work is limited to 8-bit audio, leaving open whether such approaches work for practical settings (16/24-bit) and can compete with existing codecs. We benchmark LM-based compression on full-fidelity audio across diverse domains (music, speech, bioacoustics), sampling rates (16kHz-48kHz), and bit depths (8, 16, 24-bit). Standard sample-level tokenization becomes intractable at higher bit depths due to vocabulary size (65K for 16-bit; 16.7M for 24-bit). We propose Trilobyte, a byte-level tokenization schema for full resolution audio, improving vocabulary scaling from $O(2^{b})$ to $O(1)$ and enabling the first tractable 24-bit LM-based lossless compression. While LMs consistently outperform FLAC and yield state-of-the-art compression at 8-bit and 16-bit, we observe that compression gains become more modest as bit depth increases beyond 8-bit.
Sound (cs.SD); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
arXiv:2603.08683 [cs.SD]
(or arXiv:2603.08683v3 [cs.SD] for this version)
https://doi.org/10.48550/arXiv.2603.08683 Focus to learn more arXiv-issued DOI via DataCite
Proc. Interspeech 2026, 3889-3893
https://doi.org/10.21437/Interspeech.2026-1748 Focus to learn more DOI(s) linking to related resources
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 250 cuvinte