Studies have shown that in noisy acoustic environments, providing binaural signals to the user of an assistive listening device may improve speech intelligibility and spatial awareness. This paper presents a binaural speech enhancement method using a complex convolutional neural network with an encoder-decoder architecture and a complex multi-head attention transformer. The model is trained to estimate individual complex ratio masks in the time-frequency domain for the left and right-ear channels of binaural hearing devices. The model is trained using a novel loss function that incorporates the preservation of spatial information along with speech intelligibility improvement and noise reduction. Simulation results for acoustic scenarios with a single target speaker and isotropic noise of various types show that the proposed method improves the estimated binaural speech intelligibility and preserves the binaural cues better in comparison with several baseline algorithms.
Recently, a mask-based beamformer with attention-based spatial covariance matrix aggregator (ASA) was proposed, which was demonstrated to track moving sources accurately. However, the deep neural network model used in this algorithm is limited to a specific channel configuration, requiring a different model in case a different channel permutation, channel count, or microphone array geometry is considered. Addressing this limitation, in this paper, we investigate three approaches to improve the robustness of the ASA-based tracking method against such variations: incorporating random channel configurations during the training process, employing the transform-average-concatenate (TAC) method to process multi-channel input features (allowing for any channel count and enabling permutation invariance), and utilizing input features that are robust against variations of the channel configuration. Our experiments, conducted using the CHiME-3 and DEMAND datasets, demonstrate improved robustness against mismatches in channel permutations, channel counts, and microphone array geometries compared to the conventional ASA-based tracking method without compromising performance in matched conditions, suggesting that the mask-based beamformer with ASA integrating the proposed approaches has the potential to track moving sources for arbitrary microphone arrays.
Reverberation can severely degrade the quality of speech signals recorded using microphones in an enclosure. In acoustic sensor networks with spatially distributed microphones, a similar dereverberation performance may be achieved using only a subset of all available microphones. Using the popular convex relaxation method, in this paper we propose to perform microphone subset selection for the weighted prediction error (WPE) multi-channel dereverberation algorithm by introducing a group sparsity penalty on the prediction filter coefficients. The resulting problem is shown to be solved efficiently using the accelerated proximal gradient algorithm. Experimental evaluation using measured impulse responses shows that the performance of the proposed method is close to the optimal performance obtained by exhaustive search, both for frequency-dependent as well as frequency-independent microphone subset selection. Furthermore, the performance using only a few microphones for frequency-independent microphone subset selection is only marginally worse than using all available microphones.
To estimate the direction of arrival (DOA) of multiple speakers with methods that use prototype transfer functions, frequency-dependent spatial spectra (SPS) are usually constructed. To make the DOA estimation robust, SPS from different frequencies can be combined. According to how the SPS are combined, frequency fusion mechanisms are categorized into narrowband, broadband, or speaker-grouped, where the latter mechanism requires a speaker-wise grouping of frequencies. For a binaural hearing aid setup, in this paper we propose an interaural time difference (ITD)-based speaker-grouped frequency fusion mechanism. By exploiting the DOA dependence of ITDs, frequencies can be grouped according to a common ITD and be used for DOA estimation of the respective speaker. We apply the proposed ITD-based speaker-grouped frequency fusion mechanism for different DOA estimation methods, namely the multiple signal classification, steered response power and a recently published method based on relative transfer function (RTF) vectors. In our experiments, we compare DOA estimation with different fusion mechanisms. For all considered DOA estimation methods, the proposed ITD-based speaker-grouped frequency fusion mechanism results in a higher DOA estimation accuracy compared with the narrowband and broadband fusion mechanisms.
Spatially selective active noise control (ANC) hearables are designed to reduce unwanted noise from certain directions while preserving desired sounds from other directions. In previous studies, the target signal has been defined either as the delayed desired component in one of the reference microphone signals or as the desired component in the error microphone signal without any delay. In this paper, we systematically investigate the influence of delays in different target signals on the ANC performance and provide an intuitive explanation for how the system obtains the desired signal. Simulations were conducted on a pair of open-fitting hearables for localized speech and noise sources in an anechoic environment. The performance was assessed in terms of noise reduction, signal quality and control effort. Results indicate that optimal performance is achieved without delays when the target signal is defined at the error microphone, whereas causality necessitates delays when the target signal is defined at the reference microphone. The optimal delay is found to be the acoustic delay between this reference microphone and the error microphone from the desired source.
Hearables with integrated microphones may offer communication benefits in noisy working environments, e.g. by transmitting the recorded own voice of the user. Systems aiming at reconstructing the clean and full-bandwidth own voice from noisy microphone recordings are often based on supervised learning. Recording a sufficient amount of noise required for training such a system is costly since noise transmission between outer and inner microphones varies individually. Previously proposed methods either do not consider noise, only consider noise at outer microphones or assume inner and outer microphone noise to be independent during training, and it is not yet clear whether individualized noise can benefit the training of and own voice reconstruction system. In this paper, we investigate several noise data augmentation techniques based on measured transfer functions to simulate multi-microphone noise. Using augmented noise, we train a multi-channel own voice reconstruction system. Experiments using real noise are carried out to investigate the generalization capability. Results show that incorporating augmented noise yields large benefits, in particular considering individualized noise augmentation leads to higher performance.
Determining the head orientation of a talker is not only beneficial for various speech signal processing applications, such as source localization or speech enhancement, but also facilitates intuitive voice control and interaction with smart environments or modern car assistants. Most approaches for head orientation estimation are based on visual cues. However, this requires camera systems which often are not available. We present an approach which purely uses audio signals captured with only a few distributed microphones around the talker. Specifically, we propose a novel method that directly incorporates measured or modeled speech radiation patterns to infer the talker's orientation during active speech periods based on a cosine similarity measure. Moreover, an automatic gain adjustment technique is proposed for uncalibrated, irregular microphone setups, such as ad-hoc sensor networks. In experiments with signals recorded in both anechoic and reverberant environments, the proposed method outperforms state-of-the-art approaches, using either measured or modeled speech radiation patterns.
In many multi-microphone algorithms for noise reduction, an estimate of the relative transfer function (RTF) vector of the target speaker is required. The state-of-the-art covariance whitening (CW) method estimates the RTF vector as the principal eigenvector of the whitened noisy covariance matrix, where whitening is performed using an estimate of the noise covariance matrix. In this paper, we consider an acoustic sensor network consisting of multiple microphone nodes. Assuming uncorrelated noise between the nodes but not within the nodes, we propose two RTF vector estimation methods that leverage the block-diagonal structure of the noise covariance matrix. The first method modifies the CW method by considering only the diagonal blocks of the estimated noise covariance matrix. In contrast, the second method only considers the off-diagonal blocks of the noisy covariance matrix, but cannot be solved using a simple eigenvalue decomposition. When applying the estimated RTF vector in a minimum variance distortionless response beamformer, simulation results for real-world recordings in a reverberant environment with multiple noise sources show that the modified CW method performs slightly better than the CW method in terms of SNR improvement, while the off-diagonal selection method outperforms a biased RTF vector estimate obtained as the principal eigenvector of the noisy covariance matrix.
This paper addresses the challenge of estimating the relative transfer function (RTF) vectors of multiple speakers in a noisy and reverberant environment. More specifically, we consider a scenario where two speakers activate successively. In this scenario, the RTF vector of the first speaker can be estimated in a straightforward way and the main challenge lies in estimating the RTF vector of the second speaker during segments where both speakers are simultaneously active. To estimate the RTF vector of the second speaker the so-called blind oblique projection (BOP) method determines the oblique projection operator that optimally blocks the second speaker. Instead of blocking the second speaker, in this paper we propose a covariance blocking and whitening (CBW) method, which first blocks the first speaker and applies whitening using the estimated noise covariance matrix and then estimates the RTF vector of the second speaker based on a singular value decomposition. When using the estimated RTF vectors of both speakers in a linearly constrained minimum variance beamformer, simulation results using real-world recordings for multiple speaker positions demonstrate that the proposed CBW method outperforms the conventional BOP and covariance whitening methods in terms of signal-to-interferer-and-noise ratio improvement.
Hearables often contain an in-ear microphone, which may be used to capture the own voice of its user. However, due to ear canal occlusion the in-ear microphone mostly records body-conducted speech, which suffers from band-limitation effects and is subject to amplification of low frequency content. These transfer characteristics are assumed to vary both based on speech content and between individual talkers. It is desirable to have an accurate model of the own voice transfer characteristics between hearable microphones. Such a model can be used, e.g., to simulate a large amount of in-ear recordings to train supervised learning-based algorithms aiming at compensating own voice transfer characteristics. In this paper we propose a speech-dependent system identification model based on phoneme recognition. Using recordings from a prototype hearable, the modeling accuracy is evaluated in terms of technical measures. We investigate robustness of transfer characteristic models to utterance or talker mismatch. Simulation results show that using the proposed speech-dependent model is preferable for simulating in-ear recordings compared to a speech-independent model. The proposed model is able to generalize better to new utterances than an adaptive filtering-based model. Additionally, we find that talker-averaged models generalize better to different talkers than individual models.