Search

Personalised speech enhancement (PSE), which extracts only the speech of a target user and removes everything else from a recorded audio clip, can potentially improve users' experiences of audio AI modules deployed in the wild. To support a large variety of downstream audio tasks, such as real-time ASR and audio-call enhancement, a PSE solution should operate in a streaming mode, i.e., input audio cleaning should happen in real-time with a small latency and real-time factor. Personalisation is typically achieved by extracting a target speaker's voice profile from an enrolment audio, in the form of a static embedding vector, and then using it to condition the output of a PSE model. However, a fixed target speaker embedding may not be optimal under all conditions. In this work, we present a streaming Transformer-based PSE model and propose a novel cross-attention approach that gives adaptive target speaker representations. We present extensive experiments and show that our proposed cross-attention approach outperforms competitive baselines consistently, even when our model is only approximately half the size.

Neural networks (NNs) lack measures of "reliability" estimation that would enable reasoning over their predictions. Despite the vital importance, especially in areas of human well-being and health, state-of-the-art uncertainty estimation techniques are computationally expensive when applied to resource-constrained devices. We propose an efficient framework for predictive uncertainty estimation in NNs deployed on embedded edge systems with no need for fine-tuning or re-training strategies. To meet the energy and latency requirements of these embedded platforms the framework is built from the ground up to provide predictive uncertainty based only on one forward pass and a negligible amount of additional matrix multiplications with theoretically proven correctness. Our aim is to enable already trained deep learning models to generate uncertainty estimates on resource-limited devices at inference time focusing on classification tasks. This framework is founded on theoretical developments casting dropout training as approximate inference in Bayesian NNs. Our layerwise distribution approximation to the convolution layer cascades through the network, providing uncertainty estimates in one single run which ensures minimal overhead, especially compared with uncertainty techniques that require multiple forwards passes and an equal linear rise in energy and latency requirements making them unsuitable in practice. We demonstrate that it yields better performance and flexibility over previous work based on multilayer perceptrons to obtain uncertainty estimates. Our evaluation with mobile applications datasets shows that our approach not only obtains robust and accurate uncertainty estimations but also outperforms state-of-the-art methods in terms of systems performance, reducing energy consumption (up to 28x), keeping the memory overhead at a minimum while still improving accuracy (up to 16%)., Comment: 13 pages, 6 figures

LPCNet is an efficient vocoder that combines linear prediction and deep neural network modules to keep the computational complexity low. In this work, we present two techniques to further reduce it's complexity, aiming for a low-cost LPCNet vocoder-based neural Text-to-Speech (TTS) System. These techniques are: 1) Sample-bunching, which allows LPCNet to generate more than one audio sample per inference; and 2) Bit-bunching, which reduces the computations in the final layer of LPCNet. With the proposed bunching techniques, LPCNet, in conjunction with a Deep Convolutional TTS (DCTTS) acoustic model, shows a 2.19x improvement over the baseline run-time when running on a mobile device, with a less than 0.1 decrease in TTS mean opinion score (MOS)., Comment: Interspeech 2020

Increasing demand for on-device Automatic Speech Recognition (ASR) systems has resulted in renewed interests in developing automatic model compression techniques. Past research have shown that AutoML-based Low Rank Factorization (LRF) technique, when applied to an end-to-end Encoder-Attention-Decoder style ASR model, can achieve a speedup of up to 3.7x, outperforming laborious manual rank-selection approaches. However, we show that current AutoML-based search techniques only work up to a certain compression level, beyond which they fail to produce compressed models with acceptable word error rates (WER). In this work, we propose an iterative AutoML-based LRF approach that achieves over 5x compression without degrading the WER, thereby advancing the state-of-the-art in ASR compression.

The ability to form images through hair-thin optical fibres promises to open up new applications from biomedical imaging to industrial inspection. Unfortunately, deployment has been limited because small changes in mechanical deformation (e.g. bending) and temperature can completely scramble optical information, distorting images. Since such changes are dynamic, correcting them requires measurement of the fibre transmission matrix (TM) in situ immediately before imaging. TM calibration typically requires access to both the proximal and distal facets of the fibre simultaneously, which is not feasible during most realistic usage scenarios without compromising the thin form factor with bulky distal optics. Here, we introduce a new approach to determine the TM of multi-mode fibre (MMF) or multi-core fibre (MCF) in a reflection-mode configuration without access to the distal facet. A thin stack of structured metasurface reflectors is used at the distal facet to introduce wavelength-dependent, spatially heterogeneous reflectance profiles. We derive a first-order fibre model that compensates these wavelength-dependent changes in the TM and show that, consequently, the reflected data at 3 wavelengths can be used to unambiguously reconstruct the full TM by an iterative optimisation algorithm. We then present a method for sample illumination and imaging following TM reconstruction. Unlike previous approaches, our method does not require the TM to be unitary making it applicable to physically realistic fibre systems. We demonstrate TM reconstruction and imaging first using simulated non-unitary fibres and noisy reflection matrices, then using much larger experimentally-measured TMs of a densely-packed MCF, and finally on an experimentally-measured multi-wavelength set of TMs recorded from a MMF. Our findings pave the way for online transmission matrix calibration in situ in hair-thin optical fibres, Comment: Main text: 38 pages, 9 Figures, Appendices: 26 pages, 6 Figures. Corrected author affiliation

We introduce a framework for the reconstruction of the amplitude, phase and polarisation of an optical vector-field using calibration measurements acquired by an imaging device with an unknown linear transformation. By incorporating effective regularisation terms, this new approach is able to recover an optical vector-field with respect to an arbitrary representation system, which may be different from the one used in calibration. In particular, it enables the recovery of an optical vector-field with respect to a Fourier basis, which is shown to yield indicative features of increased scattering associated with tissue abnormalities. We demonstrate the effectiveness of our approach using synthetic holographic images as well as biological tissue samples in an experimental setting where measurements of an optical vector-field are acquired by a fibre endoscope, and observe that indeed the recovered Fourier coefficients are useful in distinguishing healthy tissues from lesions in early stages of oesophageal cancer.

We consider the characterization of the asymptotic behavior of the average minimum mean-squared error (MMSE) and the average mutual information in scalar and vector fading coherent channels, where the receiver knows the exact fading channel state but the transmitter knows only the fading channel distribution, driven by a range of inputs. We construct low-snr and -- at the heart of the novelty of the contribution -- high-snr asymptotic expansions for the average MMSE and the average mutual information for coherent channels subject to Rayleigh fading, Ricean fading or Nakagami fading and driven by discrete inputs (with finite support) or various continuous inputs. We reveal the role that the so-called canonical MMSE in a standard additive white Gaussian noise (AWGN) channel plays in the characterization of the asymptotic behavior of the average MMSE and the average mutual information in a fading coherent channel. We also reveal connections to and generalizations of the MMSE dimension. The most relevant element that enables the construction of these non-trivial expansions is the realization that the integral representation of the estimation- and information- theoretic quantities can be seen as an h-transform of a kernel with a monotonic argument: this enables the use of a novel asymptotic expansion of integrals technique -- the Mellin transform method -- that leads immediately to not only the high-snr but also the low-snr expansions of the average MMSE and -- via the I-MMSE relationship -- to expansions of the average mutual information. We conclude with applications of the results to the characterization and optimization of the constrained capacity of a bank of parallel independent coherent fading channels driven by arbitrary discrete inputs.