MLMA: Towards Multilingual ASR With Mamba-based Architectures
Mohamed Nabih Ali, Daniele Falavigna, Alessio Brutti
TL;DR
MLMA addresses the challenge of scalable multilingual ASR by leveraging the Mamba state-space architecture. The approach uses a ConMamba encoder paired with a CTC decoder, trained on nearly 11.8k hours across 6 European languages, and demonstrates competitive performance against Conformer baselines and large multilingual models while offering faster inference and robust long-context processing. The results across monolingual, bilingual, and multilingual settings, including out-of-domain data like FLEURS, suggest that state-space models are viable backbones for inclusive, real-time multilingual ASR. Overall, the work positions Mamba as a scalable, efficient alternative for building broad-language ASR systems with practical deployment potential.
Abstract
Multilingual automatic speech recognition (ASR) remains a challenging task, especially when balancing performance across high- and low-resource languages. Recent advances in sequence modeling suggest that architectures beyond Transformers may offer better scalability and efficiency. In this work, we introduce MLMA (Multilingual Language Modeling with Mamba for ASR), a new approach that leverages the Mamba architecture -- an efficient state-space model optimized for long-context sequence processing -- for multilingual ASR. Using Mamba, MLMA implicitly incorporates language-aware conditioning and shared representations to support robust recognition across diverse languages. Experiments on standard multilingual benchmarks show that MLMA achieves competitive performance compared to Transformer-based architectures. These results highlight Mamba's potential as a strong backbone for scalable, efficient, and accurate multilingual speech recognition.