E. Chng
Publications
Privacy-Preserving End-to-End Full-Duplex Speech Dialogue Models
End-to-end full-duplex speech models feed user audio through an always-on LLM backbone, yet the speaker privacy implications of their hidden representations remain unexamined. Following the VoicePrivacy 2024 protocol with a lazy-informed attacker, we show that the hidden states of SALM-Duplex and Moshi leak substantial speaker identity across all transformer layers. Layer-wise and turn-wise analyses reveal that leakage persists across all layers, with SALM-Duplex showing stronger leakage in early layers while Moshi leaks uniformly, and that Linkability rises sharply within the first few turns. We propose two streaming anonymization setups using Stream-Voice-Anon: a waveform-level front-end (Anon-W2W) and a feature-domain replacement (Anon-W2F). Anon-W2F raises EER by over 3.5x relative to the discrete encoder baseline (11.2% to 41.0%), approaching the 50% random-chance ceiling, while Anon-W2W retains 78-93% of baseline sBERT across setups with sub-second response latency (FRL under 0.8 s).
Prosodic Boundary-Aware Streaming Generation for LLM-Based TTS with Streaming Text Input
Streaming TTS that receives streaming text is essential for interactive systems, yet this scheme faces two major challenges: unnatural prosody due to missing lookahead and long-form collapse due to unbounded context. We propose a prosodic-boundary-aware post-training strategy, adapting a pretrained LLM-based TTS model using weakly time-aligned data. Specifically, the model is adapted to learn early stopping at specified content boundaries when provided with limited future text. During inference, a sliding-window prompt carries forward previous text and speech tokens, ensuring bounded context and seamless concatenation. Evaluations show our method outperforms CosyVoice-Style interleaved baseline in both short and long-form scenarios. In long-text synthesis, especially, it achieves a 66.2% absolute reduction in word error rate (from 71.0% to 4.8%) and increases speaker and emotion similarity by 16.1% and 1.5% relatively, offering a robust solution for streaming TTS with incremental text.
StreamVoiceAnon+: Emotion-Preserving Streaming Speaker Anonymization via Frame-Level Acoustic Distillation
We address the challenge of preserving emotional content in streaming speaker anonymization (SA). Neural audio codec language models trained for audio continuation tend to degrade source emotion: content tokens discard emotional information, and the model defaults to dominant acoustic patterns rather than preserving paralinguistic attributes. We propose supervised finetuning with neutral-emotion utterance pairs from the same speaker, combined with frame-level emotion distillation on acoustic token hidden states. All modifications are confined to finetuning, which takes less than 2 hours on 4 GPUs and adds zero inference latency overhead, while maintaining a competitive 180ms streaming latency. On the VoicePrivacy 2024 protocol, our approach achieves a 49.2% UAR (emotion preservation) with 5.77% WER (intelligibility), a +24% relative UAR improvement over the baseline (39.7%->49.2%) and +10% over the emotion-prompt variant (44.6% UAR), while maintaining strong privacy (EER 49.0%). Demo and code are available: https://anonymous3842031239.github.io/
Training-Free Intelligibility-Guided Observation Addition for Noisy ASR
Automatic speech recognition (ASR) degrades severely in noisy environments. Although speech enhancement (SE) front-ends effectively suppress background noise, they often introduce artifacts that harm recognition. Observation addition (OA) addressed this issue by fusing noisy and SE enhanced speech, improving recognition without modifying the parameters of the SE or ASR models. This paper proposes an intelligibility-guided OA method, where fusion weights are derived from intelligibility estimates obtained directly from the backend ASR. Unlike prior OA methods based on trained neural predictors, the proposed method is training-free, reducing complexity and enhances generalization. Extensive experiments across diverse SE-ASR combinations and datasets demonstrate strong robustness and improvements over existing OA baselines. Additional analyses of intelligibility-guided switching-based alternatives and frame versus utterance-level OA further validate the proposed design.
TextBFGS: Quasi-Newton Optimization for Discrete Executable Text via Gradient-Operator Retrieval
Optimizing discrete executable text such as prompts and code has recently been framed as a gradient-based process, effectively translating backpropagation concepts to the semantic space. However, existing methods predominantly operate as first-order optimizers akin to Stochastic Gradient Descent, which are suffering from slow convergence and instability because they neglect the semantic curvature of the optimization landscape. To bridge this gap, we introduce TextBFGS, a second-order framework to implement a Quasi-Newton optimization method for discrete text. Unlike traditional memory-based approaches that retrieve similar textual instances, TextBFGS approximates the inverse Hessian matrix by retrieving Gradient-Operators from the memory of pre-learned successful trajectories. Specifically, given a textual gradient feedback, TextBFGS identifies historical correction patterns from the optimization knowledge base and tries to apply these abstract operators to the current variable. This mechanism enables a One-Pass Update, combining feedback generation and second-order correction into a single inference step. Empirical evaluations on code optimization across diverse domains (e.g., HumanEval, MBPP) demonstrate that TextBFGS significantly outperforms first-order baselines. It achieves superior pass rates with fewer model calls and exhibits strong cross-task transferability, thus establishes a mathematically grounded paradigm for efficient, memory-aware text optimization.