DRASP: A Dual-Resolution Attentive Statistics Pooling Framework for Automatic MOS Prediction

2025/08/29 by Cheng-Yeh Yang, Yang, Cheng-Yeh, Kui Huang +9
Computer Science · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Music and Audio Processing #Sound (cs.SD) #Speech Recognition and Synthesis #Speech and Audio Processing

paper · pdf · doi:10.48550/arxiv.2508.21407

openalex publication_date 2025/08/29 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

A pooling mechanism is essential for mean opinion score (MOS) prediction, facilitating the transformation of variable-length audio features into a concise fixed-size representation that effectively encodes speech quality. Existing pooling methods typically operate at a singular granularity, concentrating either on a comprehensive global perspective or a detailed frame-level analysis, which may overlook complementary perceptual insights. To address this limitation, we introduce the Dual-Resolution Attentive Statistics Pooling (DRASP) framework. DRASP integrates both coarse-grained, global statistical summaries and fine-grained, attentive analyses of perceptually significant segments. This dual-view architecture empowers our model to formulate a more thorough and robust representation, capturing both the overarching structural context and salient local details concurrently. Extensive experiments validate the effectiveness and strong generalization ability of the proposed framework. It consistently outperforms various baseline methods across diverse datasets (MusicEval and AES-Natural), MOS prediction backbones (including a CLAP-based model and AudioBox-Aesthetics), and different audio generation systems, achieving a relative improvement of 10.39% in system-level Spearman's rank correlation coefficient (SRCC) over the widely-used average pooling approach.

Citations

• Meta Audiobox Aesthetics: Unified Automatic Quality Assessment for Speech, Music, and Sound
• MusicEval: A Generative Music Dataset with Expert Ratings for Automatic Text-to-Music Evaluation
• F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching
• VALL-E 2: Neural Codec Language Models are Human Parity Zero-Shot Text to Speech Synthesizers
• Tango 2: Aligning Diffusion-based Text-to-Audio Generations through Direct Preference Optimization
• Masked Audio Generation using a Single Non-Autoregressive Transformer
• MusicLDM: Enhancing Novelty in Text-to-Music Generation Using Beat-Synchronous Mixup Strategies
• CAM++: A Fast and Efficient Network for Speaker Verification Using Context-Aware Masking
• Large-scale Contrastive Language-Audio Pretraining with Feature Fusion and Keyword-to-Caption Augmentation
• Generalization Ability of MOS Prediction Networks
• RSKNet-MTSP: Effective and Portable Deep Architecture for Speaker Verification
• Self Multi-Head Attention for Speaker Recognition
• VoiceID Loss: Speech Enhancement for Speaker Verification
• Fréchet Audio Distance: A Metric for Evaluating Music Enhancement Algorithms
• Deep Segment Attentive Embedding for Duration Robust Speaker Verification
• GhostVLAD for set-based face recognition
• Exploring the Encoding Layer and Loss Function in End-to-End Speaker and Language Recognition System
• Attention Is All You Need
• Quality Aware Network for Set to Set Recognition
• Neural Aggregation Network for Video Face Recognition
• NetVLAD: CNN architecture for weakly supervised place recognition

Related

• MOSPC: MOS Prediction Based on Pairwise Comparison
• Multi-Sampling-Frequency Naturalness MOS Prediction Using Self-Supervised Learning Model with Sampling-Frequency-Independent Layer
• Generalization Ability of MOS Prediction Networks
• MBNet: MOS Prediction for Synthesized Speech with Mean-Bias Network
• LDNet: Unified Listener Dependent Modeling in MOS Prediction for Synthetic Speech