DisSpeech: Low-Resource Controllable Mandarin Stuttered Speech Synthesis for ASR Augmentation
Authors: Yao Lu
Organizations: TMCC, College of Computer Science, Nankai University, Tianjin, China
Abstract
Stuttered speech recognition remains challenging, with disfluencies such as repetitions, prolongations, and blocks disrupting speech continuity and acoustic patterns. This problem is further aggravated in Mandarin scenarios by the limited availability of stuttered speech data, which makes it difficult to train robust ASR models for diverse disfluency patterns. To address this problem, this paper proposes DisSpeech, a discrete speech token-based framework for low-resource controllable Mandarin stuttered speech synthesis and ASR data augmentation. The proposed framework introduces explicit stuttering event labels to control different disfluency patterns. Text and stuttering event labels are mapped into semantic speech tokens by a non-autoregressive masked generative Transformer, followed by prosody-aware acoustic reconstruction with explicit pitch and energy modeling. With fine-tuning using less than 50 hours of Mandarin stuttered speech, DisSpeech can generate controllable stuttered speech with competitive speech quality. Experimental results show that the proposed method outperforms previous stuttered speech synthesis methods in both speech quality and event controllability. Furthermore, the synthesized stuttered speech effectively improves multiple ASR models, with Qwen3-ASR-0.6B achieving a state-of-the-art CER of 4.19% on the evaluated Mandarin stuttered speech recognition task, while causing only slight degradation on fluent speech.
Audio-based stuttering systems to date have been trained for detection -- what disfluency is present now -- leaving prediction, the capability needed for closed-loop intervention, unstudied at deployable scale. We train a 616K-parameter CNN on SEP-28k (Apple, 20,131 three-second clips) to predict whether the next contiguous clip contains any disfluency. (1) Severity-selective precursor signal: on the episode-grouped test set, aggregate preblock AUC is modest (0.581 [0.542, 0.619]), but stratifying by upcoming event type reveals concentration on clinically severe events -- blocks 0.601 [0.554, 0.651] and sound repetitions 0.617 [0.567, 0.667] both exclude chance, while fillers (0.45) and word repetitions (0.49) are at chance. The aggregate objective converges to a severity-selective predictor because severe events carry prosodic precursors; fillers do not. (2) Cross-population transfer: without fine-tuning, the same checkpoint applied to 1,024 pediatric Children-Who-Stutter utterances (FluencyBank Teaching) attains AUC 0.674 detection and 0.655 prediction; DisfluencySpeech and LibriStutter reach 0.58-0.60 AUC. (3) Deployable on-device: lossless export to CoreML (1.19 MB), ONNX (40 KB), TFLite. Neural-Engine latency per 3 s window: 0.25 ms (iPhone 17 Pro Max, A19 Pro) to 0.55 ms (iPhone SE 3rd-gen and M1 Max). A 4 Hz streaming simulation uses 0.54% of the real-time budget. Platt-calibrated outputs (test ECE 0.010, from 0.177 raw). Five negative ablations -- output-level Future-Guided Learning, multi-clip GRU, time-axis concatenation, asymmetric focal loss, direct block-targeted training -- none improved over the vanilla baseline.
Despite advances in large-scale Automatic Speech Recognition (ASR), disfluent speech remains challenging, as state-of-the-art systems are often optimized to omit disfluencies, leading to information loss and hallucinations. Prior work has focused on verbatim transcription and the integration of disfluency markers, but adapting models on limited datasets can lead to catastrophic forgetting of general-domain knowledge. We address this gap by leveraging continual learning (CL) with explicit disfluency tokens. We first introduce these tokens into a pretrained ASR model to establish stable token mechanisms, and then continue training on additional datasets with varying disfluency distributions. Through a detailed analysis of model dynamics during training, we identify a trade-off between marker learning and ASR performance, and a consistent cross-attention head mechanism shared across CL methods.
Henri-Leon Kordt, Theresa Pekarek Rosin, Jae Hee Lee +1
Child speech differs from adult speech in acoustics, prosody, and linguistic structures. Speech disfluencies (such as repetitions) further challenge automatic understanding. While Audio Language Models (ALMs) show strong semantic reasoning from speech audio, their ability to reason about disfluent child speech in mixed-speaker settings remains unexplored. We investigate this through two tasks: child-focused semantic summarization and speech entailment. Experiments use recordings of children who stutter in mixed speaker interviews without explicit speaker separation. Models are instruction-guided to focus on the child, preserve clinically relevant disfluencies, and avoid adult-speech leakage. Evaluation combines LLM-based judges and reference-based metrics, anchored by transcript-oracle baselines to isolate errors. Results show that while ALMs extract high-level meaning from stuttered speech, reasoning degrades significantly with increased