Is One Score Enough? Assessing Singing Quality of Songs with Temporal Score Curves

2026-07-17

AI Summary

This work addresses the limitations of existing singing quality assessment methods, which are often confined to short audio segments, struggle to model time-varying vocal performance across full-length songs, and rely on scarce segment-level annotations. To overcome these challenges, we propose SongSQA, a two-stage framework that first leverages a teacher model to generate pseudo-labels for training a segment-level quality predictor, and then dynamically aggregates segment features and scores through learnable song embeddings and self-attention mechanisms. This enables fine-grained, full-song quality evaluation without requiring manual segment annotations, while producing a temporal quality curve. Our approach is the first to introduce a dynamic aggregation mechanism that explicitly models the relationship between local performance characteristics and global song quality, achieving significant improvements over baselines across multiple datasets—with up to a 13.95% gain in Kendall’s Tau (KTAU) correlation.

Abstract

Singing Quality Assessment (SQA) has become increasingly important for practical multimedia applications and Music AI systems, yet existing studies predominantly focus on short singing clips and remain insufficient for full-length songs. Unlike clip-level assessment, full-length song SQA requires modeling how singing quality varies across different audio segments and how these local variations influence the overall evaluation of vocal performance. Moreover, the scarcity of segment-level annotations makes effective supervision challenging, as directly assigning a single overall score label to every segment tends to treat different segment qualities as equivalent. To address these challenges, we propose SongSQA, a two-stage framework for full-length song SQA. In the first stage, a Segment Score Predictor is trained with pseudo labels generated by a pre-trained teacher model, enabling segment-level singing quality prediction without requiring manual segment annotations. In the second stage, a Song Quality Aggregator integrates segment features and predicted segment scores into unified segment embeddings, and employs a learnable song embedding together with self-attention to capture the connection between segment-level vocal performance and overall song quality. In this way, SongSQA dynamically aggregates critical quality cues across the song to produce a holistic quality prediction, while also generating a temporal segment-level quality curve. Experimental results demonstrate the effectiveness of SongSQA for full-length song SQA, achieving up to a 13.95% relative improvement in KTAU over the strongest baseline, while consistently improving other evaluation metrics across all datasets.

Problem

• Singing Quality Assessment
• full-length song
• temporal score curves
• segment-level annotations
• overall quality prediction

Innovation

• Singing Quality Assessment
• Temporal Score Curve
• Segment-level Prediction
• Self-attention Aggregation
• Pseudo-labeling