Mizar, a 159.3M-parameter ALM, is introduced, a recipe that brings together architecture, data, and three-stage training to build Mizar, a 159.3M-parameter ALM that surpasses the previous best-performing ALM below 200M parameters on all three benchmarks.
Abstract
Audio-language models (ALMs) integrate acoustic perception with the knowledge encoded in language models, enabling contextual understanding of auditory events. Making these capabilities practical on devices with limited memory and computation motivates our focus on small ALMs with fewer than 200M parameters. We introduce a recipe that brings together architecture, data, and three-stage training to build Mizar, a 159.3M-parameter ALM. Its architecture connects a compact CED-Small audio encoder to SmolLM2-135M through a frequency-merging mapper. With supervision drawn from ReasonAQA, AudioMCQ, and AVQA, the model undergoes three training stages: audio-language alignment (Stage 1), audio-dependent fine-tuning (Stage 2), and post-training (Stage 3) aimed at strengthening weak skills while retaining learned capabilities. Across five random seeds, Mizar achieves mean accuracies of 52.92% on MMAU, 42.42% on MMAR, and 36.02% on ADQA-clean, surpassing the previous best-performing ALM below 200M parameters on all three benchmarks. It also supports local inference on a single CPU: on questions from the MMAU benchmark, the mean latency from opening the audio file to generating a complete answer is 1.09 seconds. Code and checkpoints are available at https://github.com/KaiyangLi1992/Mizar_159M.
Audio large language models (ALLMs) can understand speech content, yet their ability to use speaker identity for verification remains limited. We propose TS-SP (Two-Stage Speaker Preservation), a parameter-efficient framework for learning speaker-preserving representations and making them accessible to an ALLM's langua...
Jun-Jie Li, Zheng Liang, Zhe Li et al.· 0 citations
STAG is introduced, to the authors' knowledge the first post-hoc framework for token-level spectro-temporal grounding of captions generated by audio-based MLLMs, and provides behavioral support for the faithfulness and selectivity of the explanations.
Qwen-Audio-3.0-ASR is presented, a Mixture-of-Experts (MoE) LLM-based ASR system designed to address production demands through a unified, instruction-following framework and demonstrates state-of-the-art or highly competitive recognition performance across a broad range of evaluation conditions.
Chuan-Meng Bian, Da-Ren Chen, Pei-Xin Chen et al.· 3 citations
A mechanistic analysis of paralinguistic information in four open source models using the Expresso dataset with controlled speaking styles identifies a gap between what models encode and what they use, highlighting a key limitation in current audio language models.
FireRedAudio is introduced, a general-purpose audio language model with a shared 9B-parameter LLM that achieves competitive or leading performance in audio understanding and multilingual ASR, strong content accuracy and speaker preservation in zero-shot TTS, leading instruction following in Instruct TTS, and substantia...
Fei-Yu Shen, Fenglong Xie, Junjie Li et al.· 3 citations· ⚡1
MuLA-Bench exposes conditional failure patterns that a single long-context score does not capture, and evaluates ten audio-language models and conducts pooled diagnostics on a fixed eight-model cohort.
Ze-Yu Yang, Xin-Yu Zhang, Zi-Bo Bi et al.· arXiv.org· 2 citations