Drax: Speech Recognition with Discrete Flow Matching

2025-10-05

AI Summary

To address the performance bottleneck in non-autoregressive (NAR) automatic speech recognition (ASR) caused by train-inference distribution mismatch, this paper proposes Drax—the first NAR ASR framework based on discrete flow matching. Drax constructs an audio-conditioned probability flow path that explicitly models intermediate erroneous token trajectories during inference, thereby mitigating distributional shift between training and inference. Theoretically, it establishes a connection between generalization error and cumulative velocity error, providing principled guidance for model design. Drax enables fully parallel decoding and achieves recognition accuracy competitive with state-of-the-art autoregressive models on benchmarks including LibriSpeech, while substantially improving decoding efficiency. Extensive experiments validate the effectiveness and scalability of discrete flow matching for ASR tasks.

Technology Category

• Natural Language Processing: Learning & Optimization for NLP
• Machine Learning: Deep Neural Architectures and Foundation Models
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Diffusion and flow-based non-autoregressive (NAR) models have shown strong promise in large language modeling, however, their potential for automatic speech recognition (ASR) remains largely unexplored. We propose Drax, a discrete flow matching framework for ASR that enables efficient parallel decoding. To better align training with inference, we construct an audio-conditioned probability path that guides the model through trajectories resembling likely intermediate inference errors, rather than direct random noise to target transitions. Our theoretical analysis links the generalization gap to divergences between training and inference occupancies, controlled by cumulative velocity errors, thereby motivating our design choice. Empirical evaluation demonstrates that our approach attains recognition accuracy on par with state-of-the-art speech models while offering improved accuracy-efficiency trade-offs, highlighting discrete flow matching as a promising direction for advancing NAR ASR.

Problem

• Developing discrete flow matching for efficient parallel speech recognition decoding
• Addressing training-inference misalignment through audio-conditioned probability paths
• Improving accuracy-efficiency trade-offs in non-autoregressive automatic speech recognition

Innovation

• Discrete flow matching framework for ASR
• Audio-conditioned probability path design
• Parallel decoding with improved accuracy-efficiency trade-offs