CBF-AFA: Chunk-Based Multi-SSL Fusion for Automatic Fluency Assessment

TL;DR

This paper presents a chunk-based multi-self-supervised learning fusion approach for automatic fluency assessment, improving accuracy by integrating phonetic, prosodic, and linguistic features with hierarchical neural networks.

Contribution

It introduces a novel chunk-based fusion method combining multiple SSL models and hierarchical neural networks for enhanced fluency evaluation.

Findings

Improves F1-score by 2.8 points on Speechocean762

Increases Pearson correlation by 6.2 points on Speechocean762

Surpasses existing segmentation baselines in fluency assessment

Abstract

Automatic fluency assessment (AFA) remains challenging, particularly in capturing speech rhythm, pauses, and disfluencies in non-native speakers. We introduce a chunk-based approach integrating self-supervised learning (SSL) models (Wav2Vec2, HuBERT, and WavLM) selected for their complementary strengths in phonetic, prosodic, and noisy speech modeling, with a hierarchical CNN-BiLSTM framework. Speech is segmented into breath-group chunks using Silero voice activity detection (Silero-VAD), enabling fine-grained temporal analysis while mitigating over-segmentation artifacts. SSL embeddings are fused via a learnable weighted mechanism, balancing acoustic and linguistic features, and enriched with chunk-level fluency markers (e.g., speech rate, pause durations, n-gram repetitions). The CNN-BiLSTM captures local and long-term dependencies across chunks. Evaluated on Avalinguo and Speechocean762, our approach improves F1-score by 2.8 and Pearson correlation by 6.2 points over single SSL baselines on Speechocean762, with gains of 4.2 F1-score and 4.0 Pearson points on Avalinguo, surpassing Pyannote.audio-based segmentation baselines. These findings highlight chunk-based multi-SSL fusion for robust fluency evaluation, though future work should explore generalization to dialects with irregular prosody.