Weakly Guided and Autoregressive Beamformer Parameterization for Generalizable Moving Speaker Extraction in Higher-Order Ambisonics

Linear spatial filters (beamformers) enable robust, generalizable and interpretable speech enhancement with performance guarantees under ideal parameterization. Modern beamformers are often parameterized by deep neural networks, whose performance degrades in dynamic scenarios with multiple moving speakers of unknown directions. We propose a data-driven beamforming pipeline, which only requires an estimate of the target's initial direction. Building on a higher-order ambisonics representation, we show that neural temporal-spectral processing can be decoupled from linear spatial processing, and thereby achieve generalizable and array-agnostic enhancement. By incorporating autoregression into a frame-wise causal framework, we maintain consistent performance throughout fast speaker motion and long recordings. Evaluation on synthetic data demonstrates robust enhancement under challenging conditions with closely spaced and crossing speakers. Real-world recordings in a dynamic office meeting scenario complement these findings and show generalizability across varying ambisonics orders.

Repurposing a Speech Classifier for Guided Diffusion-Based Speech Generation

Classifier guidance is a way to control diffusion generation by using a noise-conditioned classifier to steer the sampling process toward a target class. One drawback of classifier guidance is that it requires two separately trained models: a classifier and a diffusion model. We therefore study a more compact alternative in which a conventionally trained speech classifier is repurposed as the backbone for diffusion generation. Starting from a frozen noise-conditioned classifier in log-Mel space, we attach a lightweight subnetwork that reuses intermediate classifier representations and train only this subnetwork under a Denoising Score Matching objective. Our work shows that a pretrained classifier can be repurposed for conditional generation, providing an appealing bridge between discriminative modeling and conditional speech synthesis resulting in high speech quality within a single-backbone model, with reduced memory footprint and computational cost.

Your U-Net Dereverberation Model is Secretly an RIR Encoder

In this work, we analyze the ability of NCSN++ U-Net based audio dereverberation models to capture global room characteristics in their intermediate representations. Through an empirical study of both a state-of-the-art diffusion-based model and a discriminative counterpart, we show that deeper layers encode structured room impulse response (RIR)-dependent embeddings. Moreover, the discriminative ability of this implicit room representation correlates with dereverberation performance across objective metrics. Motivated by this observation, we propose a training strategy that explicitly conditions the network on pre-trained RIR embeddings, obtained via self-supervised contrastive learning. Incorporating RIR conditioning improves representation quality, accelerates convergence, and enhances dereverberation performance, while significantly reducing the number of reverse diffusion steps required by the diffusion-based model during inference.

Too Good to Be True: A Study on Modern Automatic Speech Recognition for the Evaluation of Speech Enhancement

Speech enhancement (SE) systems are typically evaluated using a variety of instrumental metrics. The use of automatic speech recognition (ASR) systems to evaluate SE performance is common in literature, usually in terms of word error rate (WER). However, WER scores depend heavily on the choice of ASR system and text normalization pipeline. In this paper, we investigate how modern ASR models correlate with human recognition of enhanced speech. A listening experiment reveals that modern ASR models with large-scale noisy training and embedded language models correlate more with human WER than simpler ones, with a transducer model providing the most reliable transcriptions. Nevertheless, we also show that these models' robustness to noise and use of context can be uninformative to an acoustics-focused evaluation of enhancement performance.

Autoregressive Guidance of Deep Spatially Selective Filters using Bayesian Tracking for Efficient Extraction of Moving Speakers

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 when 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 propose and publish a novel dataset 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, challenging acoustic conditions.

A Fast Solver for Interpolating Stochastic Differential Equation Diffusion Models for Speech Restoration

Diffusion Probabilistic Models (DPMs) are a well-established class of diffusion models for unconditional image generation, while SGMSE+ is a well-established conditional diffusion model for speech enhancement. One of the downsides of diffusion models is that solving the reverse process requires many evaluations of a large Neural Network. Although advanced fast sampling solvers have been developed for DPMs, they are not directly applicable to models such as SGMSE+ due to differences in their diffusion processes. Specifically, DPMs transform between the data distribution and a standard Gaussian distribution, whereas SGMSE+ interpolates between the target distribution and a noisy observation. This work first develops a formalism of interpolating Stochastic Differential Equations (iSDEs) that includes SGMSE+, and second proposes a solver for iSDEs. The proposed solver enables fast sampling with as few as 10 Neural Network evaluations across multiple speech restoration tasks.

Adaptive Rotary Steering with Joint Autoregression for Robust Extraction of Closely Moving Speakers in Dynamic Scenarios

Latest advances in deep spatial filtering for Ambisonics demonstrate strong performance in stationary multi-speaker scenarios by rotating the sound field toward a target speaker prior to multi-channel enhancement. For applicability in dynamic acoustic conditions with moving speakers, we propose to automate this rotary steering using an interleaved tracking algorithm conditioned on the target's initial direction. However, for nearby or crossing speakers, robust tracking becomes difficult and spatial cues less effective for enhancement. By incorporating the processed recording as additional guide into both algorithms, our novel joint autoregressive framework leverages temporal-spectral correlations of speech to resolve spatially challenging speaker constellations. Consequently, our proposed method significantly improves tracking and enhancement of closely spaced speakers, consistently outperforming comparable non-autoregressive methods on a synthetic dataset. Real-world recordings complement these findings in complex scenarios with multiple speaker crossings and varying speaker-to-array distances.

Bone-conduction Guided Multimodal Speech Enhancement with Conditional Diffusion Models

Single-channel speech enhancement models face significant performance degradation in extremely noisy environments. While prior work has shown that complementary bone-conducted speech can guide enhancement, effective integration of this noise-immune modality remains a challenge. This paper introduces a novel multimodal speech enhancement framework that integrates bone-conduction sensors with air-conducted microphones using a conditional diffusion model. Our proposed model significantly outperforms previously established multimodal techniques and a powerful diffusion-based single-modal baseline across a wide range of acoustic conditions.

Real-Time Streamable Generative Speech Restoration with Flow Matching

Diffusion-based generative models have greatly impacted the speech processing field in recent years, exhibiting high speech naturalness and spawning a new research direction. Their application in real-time communication is, however, still lagging behind due to their computation-heavy nature involving multiple calls of large DNNs. Here, we present Stream.FM, a frame-causal flow-based generative model with an algorithmic latency of 32 milliseconds (ms) and a total latency of 48 ms, paving the way for generative speech processing in real-time communication. We propose a buffered streaming inference scheme and an optimized DNN architecture, show how learned few-step numerical solvers can boost output quality at a fixed compute budget, explore model weight compression to find favorable points along a compute/quality tradeoff, and contribute a model variant with 24 ms total latency for the speech enhancement task. Our work looks beyond theoretical latencies, showing that high-quality streaming generative speech processing can be realized on consumer GPUs available today. Stream.FM can solve a variety of speech processing tasks in a streaming fashion: speech enhancement, dereverberation, codec post-filtering, bandwidth extension, STFT phase retrieval, and Mel vocoding. As we verify through comprehensive evaluations and a MUSHRA listening test, Stream.FM establishes a state-of-the-art for generative streaming speech restoration, exhibits only a reasonable reduction in quality compared to a non-streaming variant, and outperforms our recent work (Diffusion Buffer) on generative streaming speech enhancement while operating at a lower latency.

Real-Time Streaming Mel Vocoding with Generative Flow Matching

The task of Mel vocoding, i.e., the inversion of a Mel magnitude spectrogram to an audio waveform, is still a key component in many text-to-speech (TTS) systems today. Based on generative flow matching, our prior work on generative STFT phase retrieval (DiffPhase), and the pseudoinverse operator of the Mel filterbank, we develop MelFlow, a streaming-capable generative Mel vocoder for speech sampled at 16 kHz with an algorithmic latency of only 32 ms and a total latency of 48 ms. We show real-time streaming capability at this latency not only in theory, but in practice on a consumer laptop GPU. Furthermore, we show that our model achieves substantially better PESQ and SI-SDR values compared to well-established not streaming-capable baselines for Mel vocoding including HiFi-GAN.