Abstract:Direction-of-arrival estimation of multiple speakers in a room is an important task for a wide range of applications. In particular, challenging environments with moving speakers, reverberation and noise, lead to significant performance degradation for current methods. With the aim of better understanding factors affecting performance and improving current methods, in this paper multi-speaker direction-of-arrival (DOA) estimation is investigated using a modified version of the local space domain distance (LSDD) algorithm in a noisy, dynamic and reverberant environment employing a wearable microphone array. This study utilizes the recently published EasyCom speech dataset, recorded using a wearable microphone array mounted on eyeglasses. While the original LSDD algorithm demonstrates strong performance in static environments, its efficacy significantly diminishes in the dynamic settings of the EasyCom dataset. Several enhancements to the LSDD algorithm are developed following a comprehensive performance and system analysis, which enable improved DOA estimation under these challenging conditions. These improvements include incorporating a weighted reliability approach and introducing a new quality measure that reliably identifies the more accurate DOA estimates, thereby enhancing both the robustness and accuracy of the algorithm in challenging environments.
Abstract:Egocentric videos provide comprehensive contexts for user and scene understanding, spanning multisensory perception to behavioral interaction. We propose Spherical World-Locking (SWL) as a general framework for egocentric scene representation, which implicitly transforms multisensory streams with respect to measurements of head orientation. Compared to conventional head-locked egocentric representations with a 2D planar field-of-view, SWL effectively offsets challenges posed by self-motion, allowing for improved spatial synchronization between input modalities. Using a set of multisensory embeddings on a worldlocked sphere, we design a unified encoder-decoder transformer architecture that preserves the spherical structure of the scene representation, without requiring expensive projections between image and world coordinate systems. We evaluate the effectiveness of the proposed framework on multiple benchmark tasks for egocentric video understanding, including audio-visual active speaker localization, auditory spherical source localization, and behavior anticipation in everyday activities.
Abstract:We introduce a simple and efficient text-controllable high-fidelity music generation and editing model. It operates on sequences of continuous latent representations from a low frame rate 48 kHz stereo variational auto encoder codec that eliminates the information loss drawback of discrete representations. Based on a diffusion transformer architecture trained on a flow-matching objective the model can generate and edit diverse high quality stereo samples of variable duration, with simple text descriptions. We also explore a new regularized latent inversion method for zero-shot test-time text-guided editing and demonstrate its superior performance over naive denoising diffusion implicit model (DDIM) inversion for variety of music editing prompts. Evaluations are conducted on both objective and subjective metrics and demonstrate that the proposed model is not only competitive to the evaluated baselines on a standard text-to-music benchmark - quality and efficiency-wise - but also outperforms previous state of the art for music editing when combined with our proposed latent inversion. Samples are available at https://melodyflow.github.io.
Abstract:Adding visual cues to audio-based speech separation can improve separation performance. This paper introduces AV-CrossNet, an audiovisual (AV) system for speech enhancement, target speaker extraction, and multi-talker speaker separation. AV-CrossNet is extended from the CrossNet architecture, which is a recently proposed network that performs complex spectral mapping for speech separation by leveraging global attention and positional encoding. To effectively utilize visual cues, the proposed system incorporates pre-extracted visual embeddings and employs a visual encoder comprising temporal convolutional layers. Audio and visual features are fused in an early fusion layer before feeding to AV-CrossNet blocks. We evaluate AV-CrossNet on multiple datasets, including LRS, VoxCeleb, and COG-MHEAR challenge. Evaluation results demonstrate that AV-CrossNet advances the state-of-the-art performance in all audiovisual tasks, even on untrained and mismatched datasets.
Abstract:The last decade has witnessed significant advancements in deep learning-based speech enhancement (SE). However, most existing SE research has limitations on the coverage of SE sub-tasks, data diversity and amount, and evaluation metrics. To fill this gap and promote research toward universal SE, we establish a new SE challenge, named URGENT, to focus on the universality, robustness, and generalizability of SE. We aim to extend the SE definition to cover different sub-tasks to explore the limits of SE models, starting from denoising, dereverberation, bandwidth extension, and declipping. A novel framework is proposed to unify all these sub-tasks in a single model, allowing the use of all existing SE approaches. We collected public speech and noise data from different domains to construct diverse evaluation data. Finally, we discuss the insights gained from our preliminary baseline experiments based on both generative and discriminative SE methods with 12 curated metrics.
Abstract:While far-field multi-talker mixtures are recorded, each speaker can wear a close-talk microphone so that close-talk mixtures can be recorded at the same time. Although each close-talk mixture has a high signal-to-noise ratio (SNR) of the wearer, it has a very limited range of applications, as it also contains significant cross-talk speech by other speakers and is not clean enough. In this context, we propose a novel task named cross-talk reduction (CTR) which aims at reducing cross-talk speech, and a novel solution named CTRnet which is based on unsupervised or weakly-supervised neural speech separation. In unsupervised CTRnet, close-talk and far-field mixtures are stacked as input for a DNN to estimate the close-talk speech of each speaker. It is trained in an unsupervised, discriminative way such that the DNN estimate for each speaker can be linearly filtered to cancel out the speaker's cross-talk speech captured at other microphones. In weakly-supervised CTRnet, we assume the availability of each speaker's activity timestamps during training, and leverage them to improve the training of unsupervised CTRnet. Evaluation results on a simulated two-speaker CTR task and on a real-recorded conversational speech separation and recognition task show the effectiveness and potential of CTRnet.
Abstract:Recent advancements in deep learning have demonstrated remarkable performance comparable to human capabilities across various supervised computer vision tasks. However, the prevalent assumption of having an extensive pool of training data encompassing all classes prior to model training often diverges from real-world scenarios, where limited data availability for novel classes is the norm. The challenge emerges in seamlessly integrating new classes with few samples into the training data, demanding the model to adeptly accommodate these additions without compromising its performance on base classes. To address this exigency, the research community has introduced several solutions under the realm of few-shot class incremental learning (FSCIL). In this study, we introduce an innovative FSCIL framework that utilizes language regularizer and subspace regularizer. During base training, the language regularizer helps incorporate semantic information extracted from a Vision-Language model. The subspace regularizer helps in facilitating the model's acquisition of nuanced connections between image and text semantics inherent to base classes during incremental training. Our proposed framework not only empowers the model to embrace novel classes with limited data, but also ensures the preservation of performance on base classes. To substantiate the efficacy of our approach, we conduct comprehensive experiments on three distinct FSCIL benchmarks, where our framework attains state-of-the-art performance.
Abstract:We present a new dataset called Real Acoustic Fields (RAF) that captures real acoustic room data from multiple modalities. The dataset includes high-quality and densely captured room impulse response data paired with multi-view images, and precise 6DoF pose tracking data for sound emitters and listeners in the rooms. We used this dataset to evaluate existing methods for novel-view acoustic synthesis and impulse response generation which previously relied on synthetic data. In our evaluation, we thoroughly assessed existing audio and audio-visual models against multiple criteria and proposed settings to enhance their performance on real-world data. We also conducted experiments to investigate the impact of incorporating visual data (i.e., images and depth) into neural acoustic field models. Additionally, we demonstrated the effectiveness of a simple sim2real approach, where a model is pre-trained with simulated data and fine-tuned with sparse real-world data, resulting in significant improvements in the few-shot learning approach. RAF is the first dataset to provide densely captured room acoustic data, making it an ideal resource for researchers working on audio and audio-visual neural acoustic field modeling techniques. Demos and datasets are available on our project page: https://facebookresearch.github.io/real-acoustic-fields/
Abstract:Self-supervised learned models have been found to be very effective for certain speech tasks such as automatic speech recognition, speaker identification, keyword spotting and others. While the features are undeniably useful in speech recognition and associated tasks, their utility in speech enhancement systems is yet to be firmly established, and perhaps not properly understood. In this paper, we investigate the uses of SSL representations for single-channel speech enhancement in challenging conditions and find that they add very little value for the enhancement task. Our constraints are designed around on-device real-time speech enhancement -- model is causal, the compute footprint is small. Additionally, we focus on low SNR conditions where such models struggle to provide good enhancement. In order to systematically examine how SSL representations impact performance of such enhancement models, we propose a variety of techniques to utilize these embeddings which include different forms of knowledge-distillation and pre-training.
Abstract:Ambisonics, a popular format of spatial audio, is the spherical harmonic (SH) representation of the plane wave density function of a sound field. Many algorithms operate in the SH domain and utilize the Ambisonics as their input signal. The process of encoding Ambisonics from a spherical microphone array involves dividing by the radial functions, which may amplify noise at low frequencies. This can be overcome by regularization, with the downside of introducing errors to the Ambisonics encoding. This paper aims to investigate the impact of different ways of regularization on Deep Neural Network (DNN) training and performance. Ideally, these networks should be robust to the way of regularization. Simulated data of a single speaker in a room and experimental data from the LOCATA challenge were used to evaluate this robustness on an example algorithm of speaker localization based on the direct-path dominance (DPD) test. Results show that performance may be sensitive to the way of regularization, and an informed approach is proposed and investigated, highlighting the importance of regularization information.