In this paper we propose a robust loudspeaker beamforming algorithm which is used to enhance the performance of voice driven applications in scenarios where the loudspeakers introduce the majority of the noise, e.g. when music is playing loudly. The loudspeaker beamformer modifies the loudspeaker playback signals to create a low-acoustic-energy region around the device that implements automatic speech recognition for a voice driven application (VDA). The algorithm utilises a distortion measure based on human auditory perception to limit the distortion perceived by human listeners. Simulations and real-world experiments show that the proposed loudspeaker beamformer improves the speech recognition performance in all tested scenarios. Moreover, the algorithm allows to further reduce the acoustic energy around the VDA device at the expense of reduced objective audio quality at the listener’s location.

Sound zone algorithms control the inputs to a loudspeaker array such that spatially distinct zones, each with separate audio content, are created. This work proposes a sound zone approach which includes a model of human auditory perception in the optimization problem designing the loudspeaker control filters. The control filters are therefore optimized directly for human experience, rather than by proxy through sound pressure, as is done in typical approaches. The proposed optimization problem features a perceptually weighted constraint on the bright zone reproduction error, which allows the user of the algorithm to specify the desired bright zone quality. The proposed method achieves 2 to 4 dB of additional acoustic contrast and is expected to yield less distracting dark-zone interference for the same perceived quality when compared to a traditional approach.

Estimation of the acoustic-scene related parameters such as relative transfer functions (RTFs) from source to microphones, source power spectral densities (PSDs) and PSDs of the late reverberation is essential and also challenging. Existing maximum likelihood estimators typically consider only subsets of these parameters and use each time frame separately. In this paper we explicitly focus on the single source scenario and first propose a joint maximum likelihood estimator (MLE) to estimate all parameters jointly using a single time frame. Since the RTFs are typically invariant for a number of consecutive time frames we also propose a joint maximum likelihood estimator (MLE) using multiple time frames which has similar estimation performance compared to a recently proposed reference algorithm called simultaneously confirmatory factor analysis (SCFA), but at a much lower complexity. Moreover, we present experimental results which demonstrate that the estimation accuracy, together with the performance of noise reduction, speech quality and speech intelligibility, of our proposed joint MLE outperform those of existing MLE based approaches that use only a single time frame.

Many multi-microphone algorithms depend on knowing the relative acoustic transfer functions (RTFs) of the individual sound sources in the acoustic scene. However, accurate joint RTF estimation for multiple sources is a challenging problem. Existing methods to jointly estimate the RTF for multiple sources have either no satisfying performance, or, suffer from a very large computational complexity. In this paper, we propose a method for robust estimation of the individual RTFs in a multi-source acoustic scenario. The presented algorithm is based on linear algebraic concepts and therefore of lower computational complexity compared to a recently presented state-of-the-art algorithm, while having a similar performance. Experimental results are presented to demonstrate the RTF estimation performance as well as the noise reduction performance when combining the estimated RTFs with a beamformer.

Acoustic room geometry estimation is often performed in ad hoc settings, i.e., using multiple microphones and sources distributed around the room, or assuming control over the excitation signals. We propose a fully convolutional network (FCN) that localizes reflective surfaces under the relaxed assumptions that (i) a compact array of only two microphones is available, (ii) emitter and receivers are not synchronized, and (iii) both the excitation signals and the impulse responses of the enclosures are unknown. Our FCN is trained in a supervised fashion to predict the likelihood of reflective surfaces at specific distances and directions-of-arrival (DOA). When a single reflective surface is present, up to 80% of real and virtual sources are detected, while this figure approaches 50% in rectangular rooms. Experiments on real-world recordings report similar accuracy as with artificially reverberated speech signals, validating the generalization capabilities of the framework.

In this paper, a broadband region-based near-field beamforming algorithm is proposed and demonstrated for acoustic applications. We use an eigenfilter structure with a minimum-energy cost function based on desired and undesired near-field regions. Robustness is thus achieved by focusing on signals generated from desired zones in space while rejecting signals from undesired zones. This construction leads to a linear matrix pencil formulated in terms of the array gain to these desired and undesired zones. We include a far-field model as part of the rejection zones that further improves performance in reverberant environments. We demonstrate the robustness of the algorithm in simulated and real scenarios.

In this paper, we present a spatio-temporal framework for multichannel acoustic modeling in enclosed spaces. Reverberation occurs when the sound field is enclosed between reflective boundaries (e.g. walls). We model the reverberated sound field by proper sampling of the (generalized) Fourier representation of the free-field sound field. We show that the spatial aliasing introduced by spectral sampling represents all the (damped) reflections. From the samples of the generalized spectrum, we compute the spatio-temporal sound field in the enclosed space with very low-complexity, of O(N log N) per measuring position, with N proportional to the reverberation time.

In this letter, a generalized Fourier transform is introduced and its corresponding generalized Poisson summation formula is derived. For discrete, Fourier based, signal processing, this formula shows that a special form of control on the periodic repetitions that occur due to sampling in the reciprocal domain is possible. The present paper is focused on the derivation and analysis of a weighted circular convolution theorem. We use this specific result to compute linear convolutions in the generalized Fourier domain, without the need of zero-padding. This results in faster, more resource-efficient computations. Other techniques that achieve this have been introduced in the past using different approaches. The newly proposed theory however, constitutes a unifying framework to the methods previously published.

In this letter, we present a method for low-complexity simulation of multichannel room impulse responses (RIRs). Low-complexity RIR methods will become inevitable in next generation communication systems having massive amounts of microphones/loudspeakers. For a room with rigid boundaries, we show that proper sampling of the free-field plenacoustic spectrum results in the solution of the wave equation at any position in the room. We show that the spatial aliasing introduced by spectral sampling represents the wall reflections. These wall reflections are usually modelled, at least in low-complexity simulation algorithms, by the creation of virtual free-field sources outside the room, an image source model commonly referred to as the image method. The image method requiresO(N3) operations per receiver position, whereas the newly proposed method requires onlyO(NlogN) operations per receiver position.