Speech Enhancement

The Room Acoustics and Speaker Distance Estimation (SDE) Challenge at ICASSP 2025 explores the effectiveness of augmented room impulse response (RIR) data for improving SDE model performance. This challenge at GenDARA involves generating RIRs to supplement sparse datasets and fine-tuning SDE models with the augmented data. We employ the open-source fast diffuse room impulse response generator (FastRIR) conditioned only on speaker and listener locations. We design a quality filter to ensure generated RIR alignment with challenge RIRs, and hyperparameter optimization is employed for model fine-tuning. Our approach reduces the mean absolute error (MAE) of the five positions from 1.66m to 0.6m for GWA rooms and from 2.18m to 0.69m for Treble rooms, with results demonstrating that the augmentation approach significantly improves estimation accuracy, particularly at medium to long distances.

We propose Speech Enhancement based on Drifting Models (DriftSE), a novel generative framework that formulates denoising as an equilibrium problem. Rather than relying on iterative sampling, DriftSE natively achieves one-step inference by evolving the pushforward distribution of a mapping function to directly match the clean speech distribution. This evolution is driven by a Drifting Field, a learned correction vector that guides samples toward the high-density regions of the clean distribution, which naturally facilitates training on unpaired data by matching distributions rather than paired samples. We investigate the framework under two formulations: a direct mapping from the noisy observation, and a stochastic conditional generative model from a Gaussian prior. Experiments on the VoiceBank-DEMAND benchmark demonstrate that DriftSE achieves high-fidelity enhancement in a single step, outperforming multi-step diffusion baselines and establishing a new paradigm for speech enhancement.

With the rapid advancement of speech generation technologies, the threat posed by speech deepfakes in real-time communication (RTC) scenarios has intensified. However, existing detection studies mainly focus on offline simulations and struggle to cope with the complex distortions introduced during RTC transmission, including unknown speech enhancement processes (e.g., noise suppression) and codec compression. To address this challenge, we present the first large-scale speech deepfake dataset tailored for RTC scenarios, termed \textit{RTCFake}, totaling approximately 600 hours. The dataset is constructed by transmitting speech through multiple mainstream social media and conferencing platforms (e.g., Zoom), enabling precise pairing between offline and online speech. In addition, we propose a phoneme-guided consistency learning (PCL) strategy that enforces models to learn platform-invariant semantic structural representations. In this paper, the RTCFake dataset is divided into training, development, and evaluation sets. The evaluation set further includes both unseen RTC platforms and unseen complex noise conditions, thereby providing a more realistic and challenging evaluation benchmark for speech deepfake detection. Furthermore, the proposed PCL strategy achieves significant improvements in both cross-platform generalization and noise robustness, offering an effective and generalizable modeling paradigm. The \textit{RTCFake} dataset is provided in the {https://huggingface.co/datasets/JunXueTech/RTCFake}.

Text-based speech editing aims to modify specific segments while preserving speaker identity and acoustic context. Existing methods rely on task-specific training, which incurs high data costs and struggles with temporal fidelity in unedited regions. Meanwhile, adapting Text-to-Speech (TTS) models often faces a trade-off between editing quality and consistency. To address these issues, we propose AST, an Adaptive, Seamless, and Training-free precise speech editing framework. Leveraging a pre-trained autoregressive TTS model, AST introduces Latent Recomposition to selectively stitch preserved source segments with newly synthesized targets. Furthermore, AST extends this latent manipulation to enable precise style editing for specific speech segments. To prevent artifacts at these edit boundaries, the framework incorporates Adaptive Weak Fact Guidance (AWFG). AWFG dynamically modulates a mel-space guidance signal, enforcing structural constraints only where necessary without disrupting the generative manifold. To fill the gap of publicly accessible benchmarks, we introduce LibriSpeech-Edit, a new and larger speech editing dataset. As existing metrics poorly evaluate temporal consistency in unedited regions, we propose Word-level Dynamic Time Warping (WDTW). Extensive experiments demonstrate that AST resolves the controllability-quality trade-off without extra training. Compared to the previous most temporally consistent baseline, AST improves consistency while reducing Word Error Rate by nearly 70%. Moreover, applying AST to a foundation TTS model reduces WDTW by 27%, achieving state-of-the-art speaker preservation and temporal fidelity.

Deep spatially selective filters achieve high-quality enhancement with real-time capable architectures for stationary speakers of known directions. To retain this level of performance in dynamic scenarios where only the speakers' initial directions are given, accurate, yet computationally lightweight tracking algorithms become necessary. Assuming a frame-wise causal processing style, temporal feedback allows for leveraging the enhanced speech signal to improve tracking performance. In this work, we investigate strategies to incorporate the enhanced signal into lightweight tracking algorithms and autoregressively guide deep spatial filters. Our proposed Bayesian tracking algorithms are compatible with arbitrary deep spatial filters. To increase the realism of simulated trajectories during development and evaluation, we develop a synthetic data generation framework based on the social force model. Results validate that the autoregressive incorporation significantly improves the accuracy of our Bayesian trackers, resulting in superior enhancement with none or only negligibly increased computational overhead. Real-world recordings complement these findings and demonstrate the generalizability of our methods to unseen acoustic conditions.

In existing Audio-Visual Speech Enhancement (AVSE) methods, objectives such as Scale-Invariant Signal-to-Noise Ratio (SI-SNR) and Mean Squared Error (MSE) are widely used; however, their correlation with perceived speech quality is often suboptimal and provides limited interpretability for optimization. This work proposes a reinforcement learning-based AVSE framework with a Large Language Model (LLM)-based interpretable reward model. An audio LLM generates natural language descriptions of enhanced speech, which are converted by a sentiment analysis model into a 1-5 rating score serving as the PPO reward for fine-tuning a pretrained AVSE model. Compared with scalar metrics, LLM-generated feedback is semantically rich and explicitly describes speech quality improvements. Experiments on the AVSEC-4 dataset show that the proposed method outperforms a supervised baseline and a DNSMOS-based RL baseline in PESQ, STOI, neural quality metrics, and subjective listening tests.

Recent advances in automatic speech recognition (ASR) and speech enhancement have strengthened the common belief that cleaner audio should lead to more accurate transcription. In this work, we examine whether this assumption holds for modern zero-shot ASR systems. We conduct a structured empirical study of SAM-Audio as a preprocessing step for zero-shot transcription with OpenAI Whisper. Five Whisper variants are evaluated on noisy Bengali and English speech datasets. On the English dataset, SAM-Audio increases the average PSNR from 32.28 dB to 35.99 dB and achieves higher PSNR for 71.84% of the utterances. However, WER and CER increase in every evaluated model-dataset configuration. On the Bengali dataset, Whisper large-v3 WER increases from 65.83% to 77.35%, while CER increases from 24.13% to 34.74%. On the English dataset, Whisper base WER increases from 10.53% to 21.66%, while CER increases from 4.48% to 12.50%. Utterance-level analysis further shows that the degradation affects a substantial portion of the evaluated samples, although its severity varies across Whisper variants. These findings demonstrate that improved signal-level quality does not necessarily lead to better zero-shot ASR performance and that denoising can reduce recognition accuracy.

We present TVBC (Time-Varying Biquad Cascade), an interpretable, low-latency speech enhancement model for real-time, on-device assistive hearing. A lightweight neural controller predicts, in real time, the coefficients of a differentiable cascade of 35 second-order IIR filters (biquads), so the model tracks non-stationary noise while keeping a fully interpretable processing chain: every spectral modification is an explicit, adjustable equalizer curve rather than an opaque `black-box' transform. Because the biquad cascade carries the signal processing, the controller can be made very small, driving the cascade with only 24k parameters at a 10.7ms algorithmic latency, within hearing-aid budgets, and running entirely on-device so that audio never leaves the device. We also expose the suppression-versus-preservation trade-off as an explicit control: it can be set during training through the loss weighting, and adjusted at inference, with no retraining, by mixing the noisy input with the denoised output. On hearing-aid metrics (HASPI/HASQI) the 24k model stays within about 0.02 of DFNet3 (2.3M parameters, almost two orders of magnitude larger) while using about 29X fewer multiply-accumulates, although larger black-box models still lead on reference metrics such as PESQ. We present TVBC as a proof of concept for a compact, interpretable, and controllable denoiser for on-device assistive hearing.

Speech enhancement performance degrades significantly in noisy environments, limiting the deployment of speech-controlled technologies in industrial settings, such as manufacturing plants. Existing speech enhancement solutions primarily rely on advanced digital signal processing techniques, deep learning methods, or complex software optimization approaches. This paper introduces a novel speech enhancement robotic platform that can reconfigure the geometry of a microphone array and adapt to changing acoustic conditions. A sixteen-microphone array is mounted on a robotic arm manipulator with seven degrees of freedom. The microphones are divided into four groups of four, including one group positioned near the end-effector. The system reconfigures the array by adjusting the manipulator joint angles to place the end-effector microphones closer to the target speaker, thereby improving the reference signal quality. This proposed system is a multimodal sensing, reconfigurable audio capture device that integrates sound source localization techniques, computer vision, inverse kinematics, minimum variance distortionless response beamformer, and time-frequency masking using a deep neural network. Experimental results suggest that this approach outperforms other traditional recording configurations, achieving a higher average scale-invariant signal-to-distortion ratio and lower average word error rate across multiple input signal-to-noise ratio conditions.

Neural audio codecs have largely promoted the application of language models (LMs) for speech applications. However, the effectiveness of autoregressive LM-based models in unifying speech enhancement (SE) tasks remains underexplored. In this work, we propose UniSE, a unified decoder-only LM-based framework to handle different SE tasks including speech restoration, target speaker extraction, and speech separation. Conditioned on input speech features, it autoregressively generates target discrete tokens, facilitating compatibility between distinct learning patterns of multiple tasks. To further optimize speech quality, we introduce a progressive reinforcement learning strategy with multiple assessment criteria. Experiments on several benchmarks show that UniSE achieves competitive performance compared to discriminative and generative baselines, demonstrating the capacity of LMs in unifying SE tasks. The code and demo are available at: https://github.com/alibaba/unified-audio/tree/main/QuarkAudio-UniSE.

Real-time audio-visual speech enhancement (AVSE) is a key enabler for immersive and interactive multimedia services, yet its performance is tightly constrained by network latency, uplink capacity, and computational delay. This paper presents the design, deployment, and evaluation of a complete cloud-edge-assisted AVSE system operating over a public 5G edge network. The system integrates CNN-based acoustic enhancement and OpenCV-based facial feature extraction with an LSTM fusion network to preserve temporal coherence, and is deployed on a Vodafone-compatible AWS Wavelength edge cloud. Through extensive stress testing, we analyze end-to-end performance under varying network load and adaptive multimedia profiles. Results show that compute placement at the network edge is critical for meeting real-time coherence constraints, and that uplink capacity is often the dominant bottleneck for interactive AVSE services. Only 5G and wired Ethernet consistently satisfied the required communication delay bound for uncompressed audio-video chunks, while aggressive compression reduced payload sizes by up to 80% with negligible perceptual degradation, enabling robust operation under constrained conditions. We further demonstrate a fundamental trade-off between processing latency and enhancement quality, where reduced model complexity lowers delay but degrades reconstruction performance in low-SNR scenarios. Our findings indicate that public 5G edge environments can sustain real-time, interactive AVSE workloads when network and compute resources are carefully orchestrated, although performance margins remain tighter than in dedicated infrastructures. The architectural insights derived from this study provide practical guidelines for the design of delay-sensitive multimedia and perceptual enhancement services on emerging 5G edge-cloud platforms.

Conventional speech enhancement (SE) aims to improve speech perception and intelligibility by suppressing noise without requiring enrollment speech as reference, whereas personalized SE (PSE) addresses the cocktail party problem by extracting a target speaker's speech using enrollment speech. While these two tasks tackle different yet complementary challenges in speech signal processing, they often share similar model architectures, with PSE incorporating an additional branch to process enrollment speech. This suggests developing a unified model capable of efficiently handling both SE and PSE tasks, thereby simplifying deployment while maintaining high performance. However, PSE performance is sensitive to variations in enrollment speech, like emotional tone, which limits robustness in real-world applications. To address these challenges, we propose two novel models, USEF-PNet and DSEF-PNet, both extending our previous SEF-PNet framework. USEF-PNet introduces a unified architecture for processing enrollment speech, integrating SE and PSE into a single framework to enhance performance and streamline deployment. Meanwhile, DSEF-PNet incorporates an unsupervised speech disentanglement approach by pairing a mixture speech with two different enrollment utterances and enforcing consistency in the extracted target speech. This strategy effectively isolates high-quality speaker identity information from enrollment speech, reducing interference from factors such as emotion and content, thereby improving PSE robustness. Additionally, we explore a long-short enrollment pairing (LSEP) strategy to examine the impact of enrollment speech duration during both training and evaluation. Extensive experiments on the Libri2Mix and VoiceBank DEMAND demonstrate that our proposed USEF-PNet, DSEF-PNet all achieve substantial performance improvements, with random enrollment duration performing slightly better.