We propose Speaker-Conditioned Serialized Output Training (SC-SOT), an enhanced SOT-based training for E2E multi-talker ASR. We first probe how SOT handles overlapped speech, and we found the decoder performs implicit speaker separation. We hypothesize this implicit separation is often insufficient due to ambiguous acoustic cues in overlapping regions. To address this, SC-SOT explicitly conditions the decoder on speaker information, providing detailed information about "who spoke when". Specifically, we enhance the decoder by incorporating: (1) speaker embeddings, which allow the model to focus on the acoustic characteristics of the target speaker, and (2) speaker activity information, which guides the model to suppress non-target speakers. The speaker embeddings are derived from a jointly trained E2E speaker diarization model, mitigating the need for speaker enrollment. Experimental results demonstrate the effectiveness of our conditioning approach on overlapped speech.
Separation of simultaneously active multiple speakers is a difficult task in environments with strong reverberation and many background noise sources. This paper uses the relative transfer matrix (ReTM), a generalization of the relative transfer function of a room, to propose a simple yet novel approach for separating concurrent speakers using noisy multichannel microphone recordings. The proposed method (i) allows multiple speech and background noise sources, (ii) includes reverberation, (iii) does not need the knowledge of the locations of speech and noise sources nor microphone locations and their relative geometry, and (iv) uses relatively small segment of recordings for training. We illustrate the speech source separation capability with improved intelligibility using a simulation study consisting of four speakers in the presence of three noise sources in a reverberant room. We also show the applicability of the method in a practical experiment in a real room.
We present a transformer-based architecture for voice separation of a target speaker from multiple other speakers and ambient noise. We achieve this by using two separate neural networks: (A) An enrolment network designed to craft speaker-specific embeddings, exploiting various combinations of audio and visual modalities; and (B) A separation network that accepts both the noisy signal and enrolment vectors as inputs, outputting the clean signal of the target speaker. The novelties are: (i) the enrolment vector can be produced from: audio only, audio-visual data (using lip movements) or visual data alone (using lip movements from silent video); and (ii) the flexibility in conditioning the separation on multiple positive and negative enrolment vectors. We compare with previous methods and obtain superior performance.
This paper addresses the challenge of speaker separation, which remains an active research topic despite the promising results achieved in recent years. These results, however, often degrade in real recording conditions due to the presence of noise, echo, and other interferences. This is because neural models are typically trained on synthetic datasets consisting of mixed audio signals and their corresponding ground truths, which are generated using computer software and do not fully represent the complexities of real-world recording scenarios. The lack of realistic training sets for speaker separation remains a major hurdle, as obtaining individual sounds from mixed audio signals is a nontrivial task. To address this issue, we propose a novel method for constructing a realistic training set that includes mixture signals and corresponding ground truths for each speaker. We evaluate this dataset on a deep learning model and compare it to a synthetic dataset. We got a 1.65 dB improvement in Scale Invariant Signal to Distortion Ratio (SI-SDR) for speaker separation accuracy in realistic mixing. Our findings highlight the potential of realistic training sets for enhancing the performance of speaker separation models in real-world scenarios.
Deep learning-based single-channel speaker separation has improved significantly in recent years largely due to the introduction of the transformer-based attention mechanism. However, these improvements come at the expense of intense computational demands, precluding their use in many practical applications. As a computationally efficient alternative with similar modeling capabilities, Mamba was recently introduced. We propose SepMamba, a U-Net-based architecture composed primarily of bidirectional Mamba layers. We find that our approach outperforms similarly-sized prominent models - including transformer-based models - on the WSJ0 2-speaker dataset while enjoying a significant reduction in computational cost, memory usage, and forward pass time. We additionally report strong results for causal variants of SepMamba. Our approach provides a computationally favorable alternative to transformer-based architectures for deep speech separation.
This paper introduces an area-based source separation method designed for virtual meeting scenarios. The aim is to preserve speech signals from an unspecified number of sources within a defined spatial area in front of a linear microphone array, while suppressing all other sounds. Therefore, we employ an efficient neural network architecture adapted for multi-channel input to encompass the predefined target area. To evaluate the approach, training data and specific test scenarios including multiple target and interfering speakers, as well as background noise are simulated. All models are rated according to DNSMOS and scale-invariant signal-to-distortion ratio. Our experiments show that the proposed method separates speech from multiple speakers within the target area well, besides being of very low complexity, intended for real-time processing. In addition, a power reduction heatmap is used to demonstrate the networks' ability to identify sources located within the target area. We put our approach in context with a well-established baseline for speaker-speaker separation and discuss its strengths and challenges.
Adding visual cues to audio-based speech separation can improve separation performance. This paper introduces AV-CrossNet, an \gls{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.
The current dominant approach for neural speech enhancement is based on supervised learning by using simulated training data. The trained models, however, often exhibit limited generalizability to real-recorded data. To address this, this paper investigates training enhancement models directly on real target-domain data. We propose to adapt mixture-to-mixture (M2M) training, originally designed for speaker separation, for speech enhancement, by modeling multi-source noise signals as a single, combined source. In addition, we propose a co-learning algorithm that improves M2M with the help of supervised algorithms. When paired close-talk and far-field mixtures are available for training, M2M realizes speech enhancement by training a deep neural network (DNN) to produce speech and noise estimates in a way such that they can be linearly filtered to reconstruct the close-talk and far-field mixtures. This way, the DNN can be trained directly on real mixtures, and can leverage close-talk and far-field mixtures as a weak supervision to enhance far-field mixtures. To improve M2M, we combine it with supervised approaches to co-train the DNN, where mini-batches of real close-talk and far-field mixture pairs and mini-batches of simulated mixture and clean speech pairs are alternately fed to the DNN, and the loss functions are respectively (a) the mixture reconstruction loss on the real close-talk and far-field mixtures and (b) the regular enhancement loss on the simulated clean speech and noise. We find that, this way, the DNN can learn from real and simulated data to achieve better generalization to real data. We name this algorithm SuperM2M (supervised and mixture-to-mixture co-learning). Evaluation results on the CHiME-4 dataset show its effectiveness and potential.
We propose mixture to mixture (M2M) training, a weakly-supervised neural speech separation algorithm that leverages close-talk mixtures as a weak supervision for training discriminative models to separate far-field mixtures. Our idea is that, for a target speaker, its close-talk mixture has a much higher signal-to-noise ratio (SNR) of the target speaker than any far-field mixtures, and hence could be utilized to design a weak supervision for separation. To realize this, at each training step we feed a far-field mixture to a deep neural network (DNN) to produce an intermediate estimate for each speaker, and, for each of considered close-talk and far-field microphones, we linearly filter the DNN estimates and optimize a loss so that the filtered estimates of all the speakers can sum up to the mixture captured by each of the considered microphones. Evaluation results on a 2-speaker separation task in simulated reverberant conditions show that M2M can effectively leverage close-talk mixtures as a weak supervision for separating far-field mixtures.
A robust multichannel speaker diarization and separation system is proposed by exploiting the spatio-temporal activity of the speakers. The system is realized in a hybrid architecture that combines the array signal processing units and the deep learning units. For speaker diarization, a spatial coherence matrix across time frames is computed based on the whitened relative transfer functions (wRTFs) of the microphone array. This serves as a robust feature for subsequent machine learning without the need for prior knowledge of the array configuration. A computationally efficient Spatial Activity-driven Speaker Diarization network (SASDnet) is constructed to estimate the speaker activity directly from the spatial coherence matrix. For speaker separation, we propose the Global and Local Activity-driven Speaker Extraction network (GLASEnet) to separate speaker signals via speaker-specific global and local spatial activity functions. The local spatial activity functions depend on the coherence between the wRTFs of each time-frequency bin and the target speaker-dominant bins. The global spatial activity functions are computed from the global spatial coherence functions based on frequency-averaged local spatial activity functions. Experimental results have demonstrated superior speaker, diarization, counting, and separation performance achieved by the proposed system with low computational complexity compared to the pre-selected baselines.
We propose a novel speech separation model designed to separate mixtures with an unknown number of speakers. The proposed model stacks 1) a dual-path processing block that can model spectro-temporal patterns, 2) a transformer decoder-based attractor (TDA) calculation module that can deal with an unknown number of speakers, and 3) triple-path processing blocks that can model inter-speaker relations. Given a fixed, small set of learned speaker queries and the mixture embedding produced by the dual-path blocks, TDA infers the relations of these queries and generates an attractor vector for each speaker. The estimated attractors are then combined with the mixture embedding by feature-wise linear modulation conditioning, creating a speaker dimension. The mixture embedding, conditioned with speaker information produced by TDA, is fed to the final triple-path blocks, which augment the dual-path blocks with an additional pathway dedicated to inter-speaker processing. The proposed approach outperforms the previous best reported in the literature, achieving 24.0 and 23.7 dB SI-SDR improvement (SI-SDRi) on WSJ0-2 and 3mix respectively, with a single model trained to separate 2- and 3-speaker mixtures. The proposed model also exhibits strong performance and generalizability at counting sources and separating mixtures with up to 5 speakers.
Speech separation involves extracting an individual speaker's voice from a multi-speaker audio signal. The increasing complexity of real-world environments, where multiple speakers might converse simultaneously, underscores the importance of effective speech separation techniques. This work presents a single-microphone speaker separation network with TF attention aiming at noisy and reverberant environments. We dub this new architecture as Separation TF Attention Network (Sep-TFAnet). In addition, we present a variant of the separation network, dubbed $ \text{Sep-TFAnet}^{\text{VAD}}$, which incorporates a voice activity detector (VAD) into the separation network. The separation module is based on a temporal convolutional network (TCN) backbone inspired by the Conv-Tasnet architecture with multiple modifications. Rather than a learned encoder and decoder, we use short-time Fourier transform (STFT) and inverse short-time Fourier transform (iSTFT) for the analysis and synthesis, respectively. Our system is specially developed for human-robotic interactions and should support online mode. The separation capabilities of $ \text{Sep-TFAnet}^{\text{VAD}}$ and Sep-TFAnet were evaluated and extensively analyzed under several acoustic conditions, demonstrating their advantages over competing methods. Since separation networks trained on simulated data tend to perform poorly on real recordings, we also demonstrate the ability of the proposed scheme to better generalize to realistic examples recorded in our acoustic lab by a humanoid robot. Project page: https://Sep-TFAnet.github.io
In this paper, we introduce a causal low-latency low-complexity approach for binaural multichannel blind speaker separation in noisy reverberant conditions. The model, referred to as Group Communication Binaural Filter and Sum Network (GCBFSnet) predicts complex filters for filter-and-sum beamforming in the time-frequency domain. We apply Group Communication (GC), i.e., latent model variables are split into groups and processed with a shared sequence model with the aim of reducing the complexity of a simple model only containing one convolutional and one recurrent module. With GC we are able to reduce the size of the model by up to 83 % and the complexity up to 73 % compared to the model without GC, while mostly retaining performance. Even for the smallest model configuration, GCBFSnet matches the performance of a low-complexity TasNet baseline in most metrics despite the larger size and higher number of required operations of the baseline.
When dealing with overlapped speech, the performance of automatic speech recognition (ASR) systems substantially degrades as they are designed for single-talker speech. To enhance ASR performance in conversational or meeting environments, continuous speaker separation (CSS) is commonly employed. However, CSS requires a short separation window to avoid many speakers inside the window and sequential grouping of discontinuous speech segments. To address these limitations, we introduce a new multi-channel framework called "speaker separation via neural diarization" (SSND) for meeting environments. Our approach utilizes an end-to-end diarization system to identify the speech activity of each individual speaker. By leveraging estimated speaker boundaries, we generate a sequence of embeddings, which in turn facilitate the assignment of speakers to the outputs of a multi-talker separation model. SSND addresses the permutation ambiguity issue of talker-independent speaker separation during the diarization phase through location-based training, rather than during the separation process. This unique approach allows multiple non-overlapped speakers to be assigned to the same output stream, making it possible to efficiently process long segments-a task impossible with CSS. Additionally, SSND is naturally suitable for speaker-attributed ASR. We evaluate our proposed diarization and separation methods on the open LibriCSS dataset, advancing state-of-the-art diarization and ASR results by a large margin.
Target speech extraction aims to extract, based on a given conditioning cue, a target speech signal that is corrupted by interfering sources, such as noise or competing speakers. Building upon the achievements of the state-of-the-art (SOTA) time-frequency speaker separation model TF-GridNet, we propose AV-GridNet, a visual-grounded variant that incorporates the face recording of a target speaker as a conditioning factor during the extraction process. Recognizing the inherent dissimilarities between speech and noise signals as interfering sources, we also propose SAV-GridNet, a scenario-aware model that identifies the type of interfering scenario first and then applies a dedicated expert model trained specifically for that scenario. Our proposed model achieves SOTA results on the second COG-MHEAR Audio-Visual Speech Enhancement Challenge, outperforming other models by a significant margin, objectively and in a listening test. We also perform an extensive analysis of the results under the two scenarios.
In reverberant conditions with multiple concurrent speakers, each microphone acquires a mixture signal of multiple speakers at a different location. In over-determined conditions where the microphones out-number speakers, we can narrow down the solutions to speaker images and realize unsupervised speech separation by leveraging each mixture signal as a constraint (i.e., the estimated speaker images at a microphone should add up to the mixture). Equipped with this insight, we propose UNSSOR, an algorithm for $\textbf{u}$nsupervised $\textbf{n}$eural $\textbf{s}$peech $\textbf{s}$eparation by leveraging $\textbf{o}$ver-determined training mixtu$\textbf{r}$es. At each training step, we feed an input mixture to a deep neural network (DNN) to produce an intermediate estimate for each speaker, linearly filter the estimates, and optimize a loss so that, at each microphone, the filtered estimates of all the speakers can add up to the mixture to satisfy the above constraint. We show that this loss can promote unsupervised separation of speakers. The linear filters are computed in each sub-band based on the mixture and DNN estimates through the forward convolutive prediction (FCP) algorithm. To address the frequency permutation problem incurred by using sub-band FCP, a loss term based on minimizing intra-source magnitude scattering is proposed. Although UNSSOR requires over-determined training mixtures, we can train DNNs to achieve under-determined separation (e.g., unsupervised monaural speech separation). Evaluation results on two-speaker separation in reverberant conditions show the effectiveness and potential of UNSSOR.
Independent Vector Analysis (IVA) is a popular extension of Independent Component Analysis (ICA) for joint separation of a set of instantaneous linear mixtures, with a direct application in frequency-domain speaker separation or extraction. The mixtures are parameterized by mixing matrices, one matrix per mixture. This means that the IVA mixing model does not account for any relationships between parameters across the mixtures/frequencies. The separation proceeds jointly only through the source model, where statistical dependencies of sources across the mixtures are taken into account. In this paper, we propose a mixing model for joint blind source extraction where the mixing model parameters are linked across the frequencies. This is achieved by constraining the set of feasible parameters to the manifold of half-length separating filters, which has a clear interpretation and application in frequency-domain speaker extraction.
Binaural speech separation in real-world scenarios often involves moving speakers. Most current speech separation methods use utterance-level permutation invariant training (u-PIT) for training. In inference time, however, the order of outputs can be inconsistent over time particularly in long-form speech separation. This situation which is referred to as the speaker swap problem is even more problematic when speakers constantly move in space and therefore poses a challenge for consistent placement of speakers in output channels. Here, we describe a real-time binaural speech separation model based on a Wavesplit network to mitigate the speaker swap problem for moving speaker separation. Our model computes a speaker embedding for each speaker at each time frame from the mixed audio, aggregates embeddings using online clustering, and uses cluster centroids as speaker profiles to track each speaker throughout the long duration. Experimental results on reverberant, long-form moving multitalker speech separation show that the proposed method is less prone to speaker swap and achieves comparable performance with u-PIT based models with ground truth tracking in both separation accuracy and preserving the interaural cues.
A three-stage approach is proposed for speaker counting and speech separation in noisy and reverberant environments. In the spatial feature extraction, a spatial coherence matrix (SCM) is computed using whitened relative transfer functions (wRTFs) across time frames. The global activity functions of each speaker are estimated from a simplex constructed using the eigenvectors of the SCM, while the local coherence functions are computed from the coherence between the wRTFs of a time-frequency bin and the global activity function-weighted RTF of the target speaker. In speaker counting, we use the eigenvalues of the SCM and the maximum similarity of the interframe global activity distributions between two speakers as the input features to the speaker counting network (SCnet). In speaker separation, a global and local activity-driven network (GLADnet) is used to extract each independent speaker signal, which is particularly useful for highly overlapping speech signals. Experimental results obtained from the real meeting recordings show that the proposed system achieves superior speaker counting and speaker separation performance compared to previous publications without the prior knowledge of the array configurations.
We consider the task of region-based source separation of reverberant multi-microphone recordings. We assume pre-defined spatial regions with a single active source per region. The objective is to estimate the signals from the individual spatial regions as captured by a reference microphone while retaining a correspondence between signals and spatial regions. We propose a data-driven approach using a modified version of a state-of-the-art network, where different layers model spatial and spectro-temporal information. The network is trained to enforce a fixed mapping of regions to network outputs. Using speech from LibriMix, we construct a data set specifically designed to contain the region information. Additionally, we train the network with permutation invariant training. We show that both training methods result in a fixed mapping of regions to network outputs, achieve comparable performance, and that the networks exploit spatial information. The proposed network outperforms a baseline network by 1.5 dB in scale-invariant signal-to-distortion ratio.