Calibration-Reasoning Framework for Descriptive Speech Quality Assessment
This work addresses the need for more interpretable and fine-grained speech quality assessment for audio processing applications, representing a novel method for a known bottleneck.
The paper tackles the problem of explainable speech quality assessment by introducing a post-training method that calibrates an Audio Large Language Model for multidimensional reasoning and artifact detection, achieving a state-of-the-art 0.71 mean PCC score on the QualiSpeech benchmark and a 13% improvement in MOS prediction.
Explainable speech quality assessment requires moving beyond Mean Opinion Scores (MOS) to analyze underlying perceptual dimensions. To address this, we introduce a novel post-training method that tailors the foundational Audio Large Language Model for multidimensional reasoning, detection and classification of audio artifacts. First, a calibration stage aligns the model to predict predefined perceptual dimensions. Second, a reinforcement learning stage leverages Group Relative Policy Optimization (GRPO) with dimension-specific rewards to heavily enhance accuracy of descriptions and temporal localization of quality issues. With this approach we reach state-of-the-art results of 0.71 mean PCC score on the multidimensional QualiSpeech benchmark and 13% improvement in MOS prediction driven by RL-based reasoning. Furthermore, our fine-grained GRPO rewards substantially advance the model's ability to pinpoint and classify audio artifacts in time.