Fourier Acoustics Sound Radiation And Nearfield Acoustical Holography

This paper is a review of the evolutionary history of deep learning models. It covers from the genesis of neural networks when associationism modeling of the brain is studied, to the models that dominate the last decade of research in deep learning like convolutional neural networks, deep belief networks, and recurrent neural networks. In addition to a review of these models, this paper primarily focuses on the precedents of the models above, examining how the initial ideas are assembled to construct the early models and how these preliminary models are developed into their current forms. Many of these evolutionary paths last more than half a century and have a diversity of directions. For example, CNN is built on prior knowledge of biological vision system; DBN is evolved from a trade-off of modeling power and computation complexity of graphical models and many nowadays models are neural counterparts of ancient linear models. This paper reviews these evolutionary paths and offers a concise thought flow of how these models are developed, and aims to provide a thorough background for deep learning. More importantly, along with the path, this paper summarizes the gist behind these milestones and proposes many directions to guide the future research of deep learning.

On the Origin of Deep Learning

In hearing aids, the presence of babble noise degrades hearing intelligibility of human speech greatly. However, removing the babble without creating artifacts in human speech is a challenging task in a low SNR environment. Here, we sought to solve the problem by finding a `mapping' between noisy speech spectra and clean speech spectra via supervised learning. Specifically, we propose using fully Convolutional Neural Networks, which consist of lesser number of parameters than fully connected networks. The proposed network, Redundant Convolutional Encoder Decoder (R-CED), demonstrates that a convolutional network can be 12 times smaller than a recurrent network and yet achieves better performance, which shows its applicability for an embedded system: the hearing aids.

A Fully Convolutional Neural Network for Speech Enhancement

Recent advances in wireless systems have demonstrated the possibility of tracking a person's respiration using the RF signals that bounce off her body. The resulting breathing signal can be used to infer the person's sleep quality and stages; it also allows for monitoring sleep apnea and other sleep disordered breathing (SDB); all without any body contact. Unfortunately however past work fails when people are close to each other, e.g., a couple sharing the same bed. In this case, the breathing signals of nearby individuals interfere with each other and super-impose in the received signal. This paper presents DeepBreath, the first RF-based respiration monitoring system that can recover the breathing signals of multiple individuals even when they are separated by zero distance. To design DeepBreath, we model interference due to multiple reflected RF signals and demonstrate that the original breathing can be recovered via independent component analysis (ICA). We design a full system that eliminates interference and recovers the original breathing signals. We empirically evaluate DeepBreath using 21 nights of sleep and over 150 hours of data from 13 couples who share the bed. Our results show that DeepBreath is very accurate. Specifically, the differences between the breathing signals it recovers and the ground truth are on par with the difference between the same breathing signal measured at the person's chest and belly.

Extracting Multi-Person Respiration from Entangled RF Signals

Previous research on audio source separation based on deep neural networks (DNNs) mainly focuses on estimating the magnitude spectrum of target sources and typically, phase of the mixture signal is combined with the estimated magnitude spectra in an ad-hoc way. Although recovering target phase is assumed to be important for the improvement of separation quality, it can be difficult to handle the periodic nature of the phase with the regression approach. Unwrapping phase is one way to eliminate the phase discontinuity, however, it increases the range of value along with the times of unwrapping, making it difficult for DNNs to model. To overcome this difficulty, we propose to treat the phase estimation problem as a classification problem by discretizing phase values and assigning class indices to them. Experimental results show that our classificationbased approach 1) successfully recovers the phase of the target source in the discretized domain, 2) improves signal-todistortion ratio (SDR) over the regression-based approach in both speech enhancement task and music source separation (MSS) task, and 3) outperforms state-of-the-art MSS.

PhaseNet: Discretized Phase Modeling with Deep Neural Networks for Audio Source Separation.

We propose a new blind spatial subtraction array (BSSA) consisting of a noise estimator based on independent component analysis (ICA) for efficient speech enhancement. In this paper, first, we theoretically and experimentally point out that ICA is proficient in noise estimation under a non-point-source noise condition rather than in speech estimation. Therefore, we propose BSSA that utilizes ICA as a noise estimator. In BSSA, speech extraction is achieved by subtracting the power spectrum of noise signals estimated using ICA from the power spectrum of the partly enhanced target speech signal with a delay-and-sum beamformer. This ldquopower-spectrum-domain subtractionrdquo procedure enables better noise reduction than the conventional ICA with estimation-error robustness. Another benefit of BSSA architecture is ldquopermutation robustness". Although the ICA part in BSSA suffers from a source permutation problem, the BSSA architecture can reduce the negative affection when permutation arises. The results of various speech enhancement test reveal that the noise reduction and speech recognition performance of the proposed BSSA are superior to those of conventional methods.

Blind Spatial Subtraction Array for Speech Enhancement in Noisy Environment

Effective denoising of noise-corrupted speech signals remains a challenging problem. Existing solutions typically employ some combination of noise estimation and noise elimination, either by subtraction or by filtering. The estimation of noise and the denoising are generally treated as independent aspects of the problem. In this paper we propose a new neural-network-based approach for de-noising of speech signals. The approach integrates noise estimation and denoising into a single network design, while maintaining many of the aspects of conventional noise estimation and signal denoising through a recurrent gated structure. The network thus operates as a single integrated process that can be trained to jointly estimate noise and denoise the speech signal with minimal artifacts. Noise reduction experiments on noisy speech, both with digitally added synthetic noise and real car noise, show that the proposed algorithm can recover much of the degradation caused by the noise.

Complex recurrent neural networks for denoising speech signals

An information processing apparatus and method provide logic for processing information. In one implementation, an information processing apparatus may include a receiving unit configured to receive an audio signal associated with a motion of a human mandible over a time period. The information processing apparatus may also include a determination unit configured to determine whether the motion of the human mandible corresponds to mastication, based on at least a power of the received audio signal during the time period.

Mastication detection device and mastication detection method

An audio signal processing device includes a first microphone configured to pick up audio and output a first audio signal; a second microphone configured to pick up the audio and output a second audio signal; a first frequency converter configured to convert the first audio signal to a first audio spectrum signal; a second frequency converter configured to convert the second audio signal to a second audio spectrum signal; an operating sound estimating unit configured to estimate, based on the correlation between a sound emitting member that emits an operating sound and the first and second microphones, an operating sound spectrum signal indicating the operating sound, by calculating the first and second audio spectrum signals; and an operating sound reducing unit configured to reduce the estimated operating sound spectrum signal from the first and second audio spectrum signals.

Audio signal processing device, audio signal processing method, and program

In this paper, we propose a new supervised monaural source separation based on autoencoders. We employ the autoencoder for the dictionary training such that the nonlinear network can encode the target source with high expressiveness. The dictionary is trained by each target source without the mixture signal, which makes the system independent from the context where the dictionaries will be used. In separation process, the decoder portions of the trained autoencoders are used as dictionaries to find the activations in a iterative manner such that a summation of the decoder outputs approximates the original mixture. The results of the instruments source separation experiments revealed that the separation performance of the proposed method was superior to that of the NMF.

Keiichi Osako

Papers

Blind Spatial Subtraction Array for Speech Enhancement in Noisy Environment

Complex recurrent neural networks for denoising speech signals

Mastication detection device and mastication detection method

Audio signal processing device, audio signal processing method, and program

Supervised monaural source separation based on autoencoders