Microphone Channel Research Articles

Research on Sound Source Localization of Multiple Fixed Targets Based on Machine Learning and Distributed Arrays

This paper studies a sound source localization method of multiple fixed targets based on machine learning and distributed arrays. In an outdoor open field, a three-line array was applied to collect array data and calculate latency characteristics. Then multiple classification models were established and trained. Finally, the locations of the sound source points were predicted by those models, in which the support vector machine (SVM), the nearest node (KNN), and the naive Bayesian model achieved 100% localization accuracy. Compared to the conventional method, this method has three significant advantages: First, it does not rely on the microphone channel order and does not need to be calibrated in advance, which simplifies the localization process; Second, it can fulfill high accuracy requirements, especially suitable for the scene of multiple fixed targets; Third, it has the advantage of incremental learning, as the times of localization rises, the training set is continuously enriched and the localization results become more precise.

Speech enhancement using dual microphone adaptive noise canceller

Subject of research. The subject of the study is the issue of rational placement of microphones of the main and reference channels of a two-channel adaptive noise canceller (ANC) for speech enhancement in presence of acoustic interference from a point source. An increase of the signal-to-noise ratio (SNR) at the output of the ANC relative to the input value of the main channel was chosen as a criterion for rational placement. The limitations of the model of direct propagation of signals in reverberation conditions are considered. Methods. The estimation of the SNR at the inputs of the main and reference channels is obtained on the basis of a geometric model of point sources of signal and interference and on the analytical expressions for transfer functions between signal and interference sources and microphones of the main and reference channels for a planar model of signal propagation in a free field. The main results. The conditions for microphones placement of dual channel ANC, increasing SNR, are determined. The conditions for reference channel microphone placement at a given position of main channel microphone are determined. Limitations of ANC efficiency under reverberation conditions are investigated. It is shown that the microphone of the reference channel should be located near the interference source. For large rooms, the use of dual-microphone arrays is proposed and possible signal processing algorithms (conditions of their application) are considered. To use a two-element microphone array, additional information about the position of the target source is required. Practical significance. The obtained results can be useful for distant speech acquisition in the presence of acoustic interference.

A Novel Method for Microphone Channel Frequency Response Calibration Based on Newton Algorithm

Microphone arrays are widely applied in practical applications. In an array, each channel conventionally contains an acoustic sensor and a following signal conditioning circuit (SCC). Frequency response mismatches among channels will adversely affect array’s overall performance. Existing calibration methods provide promising results; however, their accuracy varies considerably across full frequency band and the passband bandwidth shrinkage problem also arises. To address these problems, we propose a novel calibration method based on Newton algorithm. The proposed method cascades calibration filters after channel output signals allowing the frequency response calibration to be equivalent to the optimization of calibration filter coefficients. First, all uncalibrated channels’ bandwidths are analyzed using a chirp calibration signal, resulting in appropriate construction of the desired output. Next, a spectrum mean square error criterion between each channel output and the desired signal is utilized to formulate the frequency response mismatch. Finally, we employ Newton algorithm to design the calibration filter coefficients corresponding to each channel. Explicit mathematical derivation is provided. Since the cost function is quadratic, expressions of the constant Hessian matrix and its inverse are derived. Furthermore, the inverse Hessian can be efficiently computed using spectral coefficients of channel outputs. Simulation results reveal that the proposed method significantly outperforms the state-of-the-art calibration approaches in terms of calibration accuracy, calibrated passband bandwidth performance, and convergence rate. Real-world experiments also verify its effectiveness.

Macriaphone: A Stargazer’s microphone design

A highly directional, single channel microphone design is presented. The design utilizes a line of sight waveguide similar to a Keplerian reflector-type telescope. The key feature is that the waveguide is acoustically reflection-less thereby preventing off-axis reflected and diffracted fields from contaminating the measurement. The signal in the desired direction is amplified by a parabolic reflector, which terminates the waveguide and supports a single microphone at its focal point. The resulting directivity index is superior to that of a parabolic reflector microphone. Due to the line-of-sight cone provided by the waveguide, this device is particularly well suited for measuring distant sound, which is, why it is called the macriaphone (meaning far away sound in Greek). Theoretical background, numerical predictions, and experimental results from the prototype are presented.

On Training Road Surface Classifiers by Data Augmentation

It is demonstrated that data augmentation is a promising approach to reduce the size of the captured dataset required for training automatic road surface classifiers. The context is on-board systems for autonomous or semi-autonomous driving assistance: automatic power-assisted steering. Evidence is obtained by extensive experiments involving multiple captures from a 10-channel multisensor deployment: three channels from the accelerometer (acceleration in the X, Y, and Z axes); three microphone channels; two speed channels; and the torque and position of the handwheel. These captures were made under different settings: three worm-gear interface configurations; hands on or off the wheel; vehicle speed (constant speed of 10, 15, 20, 30 km/h, or accelerating from 0 to 30 km/h); and road surface (smooth flat asphalt, stripes, or cobblestones). It has been demonstrated in the experiments that data augmentation allows a reduction by an approximate factor of 1.5 in the size of the captured training dataset.

Distance Estimation Using Self-Induced Noise of an Aerial Vehicle

In this letter, we propose an algorithm to estimate the distance between an aerial vehicle and a large obstacle using the self-induced noise present during the vehicle's normal operation. We demonstrate the feasibility of using the proposed estimation method in real-time as a feedback mechanism to actively control the altitude of a blimp-like vehicle. The method is built upon a physics-based acoustic model of an unknown source emitting sound near an acoustically reflective surface. By placing two microphones beneath a motor-propeller system used to control the altitude of the vehicle, a real-time processing algorithm of the audio signals is presented that can accurately detect the distance from the microphones to the ground. The method is based upon computing the cross-correlation matrix of the signals on the two microphone channels. We develop a novel cross-correlation processor that is capable of filtering out the unknown source term and extracting the relevant time-delay representing the time it takes for an acoustic wave traveling from the microphone to the ground and back. Furthermore, we show that the method is also robust to the more complex noise source generated by a quadrotor in a static, tethered experiment. To the best of our knowledge, this is the first framework and demonstration of obstacle sensing onboard an aerial vehicle using only the self-generated noise.

Distributed Frequency Response Calibration Based on Consensus Strategy in Microphone Array

Frequency response mismatches among microphones inevitably result in performance deterioration of signal processing algorithms for microphone arrays. A frequency response calibration method based on the consensus strategy is proposed in this article. Adaptive calibration filters concatenated with microphone channels are first introduced to compensate for the frequency response mismatches among microphones. Then, the frequency response mismatch problem is formulated as the minimization of a cost function measuring the spectrum mismatches among neighboring microphone channel outputs. Finally, the consensus strategy is used to design the calibration filter in each microphone channel. By using the consensus strategy, the proposed frequency response calibration method is robust to the room impulse response and noise differences among microphones. The proposed method outperforms gain calibration methods and the centralized frequency response calibration method. Experimental results reveal the validity of the proposed method.

Towards Robust Speech Super-resolution.

Speech super-resolution (SR) aims to increase the sampling rate of a given speech signal by generating high-frequency components. This paper proposes a convolutional neural network (CNN) based SR model that takes advantage of information from both time and frequency domains. Specifically, the proposed CNN is a time-domain model that takes the raw waveform of low-resolution speech as the input, and outputs an estimate of the corresponding high-resolution waveform. During the training stage, we employ a cross-domain loss to optimize the network. We compare our model with several deep neural network (DNN) based SR models, and experiments show that our model outperforms existing models. Furthermore, the robustness of DNN-based models is investigated, in particular regarding microphone channels and downsampling schemes, which have a major impact on the performance of DNN-based SR models. By training with proper datasets and preprocessing, we improve the generalization capability for untrained microphone channels and unknown downsampling schemes.

Multichannel speaker interference reduction using frequency domain adaptive filtering

Microphone leakage or crosstalk is a common problem in multichannel close-talk audio recordings (e.g., meetings or live music performances), which occurs when a target signal does not only couple into its dedicated microphone, but also in all other microphone channels. For further signal processing such as automatic transcription of a meeting, a multichannel speaker interference reduction is required in order to eliminate the interfering speech signals in the microphone channels. The contribution of this paper is twofold: First, we consider multichannel close-talk recordings of a three-person meeting scenario with various different crosstalk levels. In order to eliminate the crosstalk in the target microphone channel, we extend a multichannel Wiener filter approach, which considers all individual microphone channels. Therefore, we integrate an adaptive filter method, which was originally proposed for acoustic echo cancellation (AEC), in order to obtain a well-performing interferer (noise) component estimation. This results in an improved speech-to-interferer ratio by up to 2.7 dB at constant or even better speech component quality. Second, since an AEC method requires typically clean reference channels, we investigate and report findings why the AEC algorithm is able to successfully estimate the interfering signals and the room impulse responses between the microphones of the interferer and the target speakers even though the reference signals are themselves disturbed by crosstalk in the considered meeting scenario.

Audio-visual based non-line-of-sight sound source localization: A feasibility study

This paper examines the feasibility of using an audio-visual methodology for sound source localization of acoustic sources hidden from direct view. A four channel microphone array is used in conjunction with LiDAR and 2D/3D mapping to merge estimated angles of arrival with room parameters for sound source localization. The time difference of arrival (TDOA) approach is used to estimate the angle(s) of arrival from an unknown sound source and a reverse ray-tracing approach is used to identify a possible source location behind a barrier/obstacle. Using a single-source zone approach based TDOA algorithm, a variety of estimated source directions are identified, each arising from multiple sound sources due to reflections in the room. Results are combined with a three-dimensional map of the space acquired from a ground robot equipped with a 2D SICK LMS LiDAR, and a reverse ray tracing approach is used to triangulate the likely position of the source. Tests were performed in both controlled and uncontrolled environments and the method was capable of finding the hidden source to within 0.5 m accuracy, which is approximately 10% of the length of an automobile. It is proposed that a methodology like this could be used to add sound source localization capabilities on autonomous vehicles to detect the position of emergency/warning sounds in traffic that may be shielded from the direct field of view.

Adaptive Frequency Response Calibration Method for Microphone Arrays

Frequency response mismatch among microphones inevitably leads to performance deterioration of signal processing algorithms of microphone arrays, especially for distributed microphone networks. An adaptive frequency response calibration method for microphone arrays based on frequency domain least mean square (FLMS) algorithm is proposed in this paper. Adaptive calibration filters are first introduced to calibrate the frequency response mismatch among microphones, which are concatenated with microphone channels. Then, the frequency response mismatch problem is formulated with a combined spectrum mean square error criterion of pairwise microphone outputs, and an unit-norm constraint is imposed to avoid undesired solutions. Finally, FLMS algorithm is used to design the calibration filter in each microphone channel. This method achieves good frequency response calibration performance even under severe ambient noise and reverberation conditions. Experimental results reveal the validity of the proposed method.

Mixture linear prediction Gammatone Cepstral features for robust speaker verification under transmission channel noise

In this paper, we present a Mixture Linear Prediction based approach for robust Gammatone Cepstral Coefficients extraction (MLPGCCs). The proposed method provides performance improvement of Automatic Speaker Verification (ASV) using i-vector and Gaussian Probabilistic Linear Discriminant Analysis GPLDA modeling under transmission channel noise. The performance of the extracted MLPGCCs was evaluated using the NIST 2008 database where a single channel microphone recorded conversational speech. The system is analyzed in the presence of different channel transmission noises such as Additive White Gaussian (AWGN) and Rayleigh fading at various Signals to Noise Ratio (SNR) levels. The evaluation results show that the MLPGCCs features are a promising way for the ASV task. Indeed, the speaker verification performance using the MLPGCCs proposed features is significantly improved compared to the conventional Gammatone Frequency Cepstral Coefficients (GFCCs) and Mel Frequency Cepstral Coefficients (MFCCs) features. For speech signals corrupted with AWGN noise at SNRs ranging from (-5 dB to 15 dB), we obtain a significant reduction of the Equal Error Rate (EER) ranging from 9.41% to 6.65% and 3.72% to 1.50%, compared with conventional MFCCs and GFCCs features respectively. In addition, when the test speech signals are corrupted with Rayleigh fading channel we achieve an EER reduction ranging from 23.63% to 7.8% and from 10.88% to 6.8% compared with conventional MFCCs and GFCCs, respectively. We also found that the combination of GFCCs and MLPGCCs gives the highest performance of speaker verification system. The best performance combination achieved is around EER from 0.43% to 0.59% and 1.92% to 3.88%.

Sound Absorption Characteristics of Sandwich Panel Made from Double Leaf Micro-perforated Panel And Natural Fiber

One of sound-absorbent developed nowadays, micro-perforated panel (MPP), effectively reduce reflected acoustic energy after approach the panel based on a Helmholtz-type resonance absorber. Some MPPs modifications have been applied to increase its performance, like the application of double-leaf MPP. Furthermore, the implementation of sound absorption materials made of natural fibers recently has invited more attention to some researcher because it is biodegradable, environmentally friendly, and economical. In this paper, a sandwich panel made of double MPP and natural fiber as sound absorptive material is investigated. The MPPs are made of a transparent acrylic board 1.5 mm thickness, 0.8 mm hole’s diameter with five mm hole distance in front panel, and five mm in the backing panel. Two different types of natural fibers panel; pineapple leaf fiber and oil palm fiber from empty fruit bunches are used in this sandwich. a low-cost impedance tube with double channel microphone is used to measure the sound absorption coefficients. It is found that the sandwich model changes the sound absorption characteristics of MPP by shifting the maximum absorption coefficient into the lower frequency and raising the absorption coefficient of the double MPP following its natural fiber panel absorption coefficient. The total impedance of sandwich gives the effect of decreasing the frequency peak of the absorption, and higher resistance of the fiber panel will increase the absorption coefficient and broaden the frequency peak.

A Smartphone-Only Pulse Transit Time Monitor Based on Cardio-Mechanical and Photoplethysmography Modalities.

This paper reports a system for monitoring pulse transit time (PTT). Using an Android smartphone and a customized sensing circuit, the system collects seismo-cardiogram (SCG), gyro-cardiogram (GCG), and photoplethysmogram (PPG) recordings. There is no need for any other external stand-alone systems. The SCG and GCG signals are recorded with the inertial sensors of the smartphone, while the PPG signal is recorded using a sensing circuit connected to the audio jack of the phone. The sensing circuit is battery-less, powered by the audio output of the smartphone using an energy harvester that converts audio tones into DC power. PPG waveforms are sampled via the microphone channel. A signal processing framework is developed and the system is experimentally verified on twenty healthy subjects at rest. The PTT is measured as the time difference between the aortic valve (AO) opening points in SCG or GCG and the fiducial points in PPG. The root-mean-square errors between the results from a stand-alone sensor system and the proposed system report 3.9 ms from SCG-based results and 3.4 ms from GCG-based results. The detection rates report more than 97.92% from both SCG and GCG results. This performance is comparable with stand-alone sensor nodes at a much lower cost.

A Battery-Less Portable ECG Monitoring System With Wired Audio Transmission.

This paper proposes a batteryless electrocardiogram (ECG) monitoring chip and intelligent system with wired audio transmission, which can be applied to long-term and real-time ECG monitoring. The ECG signal is modulated and amplified in the front-end chip and then is transmitted through the microphone channel of the 3.5-mm headphone cable. The sinusoidal signals from the left and right channels are converted into a dc supply and a precise local oscillator signal in the front-end chip, respectively. The proposed system samples the signal through the high-precision audio analog-to-digital converter in smart devices and processes it through the internal software. Therefore, no external battery, local oscillator, and complicate modules are required in this system. A chopper/amplitude modulation (AM) reused mixer amplifier is proposed, which combines chopping with AM and reuses the local oscillator signal. The closed-loop gain of the proposed amplifier varies between 20 and 47 dB and can be automatically adjusted according to the amplitude of the output signal. The proposed chip was implemented in a standard 0.18-μm CMOS process and occupies 1.07 × 0.95 mm2 of core area. The power management module outputs a 1.5-V dc voltage to power the system. When connected with the headphone cable load, the chip consumes a total power of 156 μW. The proposed monitoring system has been verified by the audio device and MATLAB in the PC. The test results show that the input reference noise of the demodulated signal is 2.12 μVrms (0.1-200Hz) and the total harmonic distortion is 0.56%@3 mV input.

Detecting Parkinson’s disease with sustained phonation and speech signals using machine learning techniques

This study investigates the processing of voice signals for detecting Parkinson’s disease. This disease is one of the neurological disorders that affect people in the world most. The approach evaluates the use of eighteen feature extraction techniques and four machine learning methods to classify data obtained from sustained phonation and speech tasks. Phonation relates to the vowel /a/ voicing task and speech to the pronunciation of a short sentence in Lithuanian language. The audio tasks were recorded using two microphone channels from acoustic cardioid (AC) and a smartphone (SP), thus allowing to evaluate the performance for different types of microphones. Five metrics were employed to analyze the classification performance: Equal Error Rate (EER) and Area Under Curve (AUC) measures from Detection Error Tradeoff (DET) and Receiver Operating Characteristic curves, Accuracy, Specificity, and Sensitivity. We compare this approach with other approaches that use the same data set. We show that the task of phonation was more efficient than speech tasks in the detection of disease. The best performance for the AC channel achieved an accuracy of 94.55%, AUC 0.87, and EER 19.01%. When using the SP channel, we have achieved an accuracy of 92.94%, AUC 0.92, and EER 14.15%.

Characterization of digital MEMS microphone elements for usage in a hostile fire detection system on a multi-rotor drone

Acoustic based Hostile Fire Detection Systems (HFDS) depend on multi-element arrays of microphones which are simultaneously sampled at sample rates often in the 10–50 ksps range. A driving component of the electronic hardware complexity of the system is the signal buffering, filtering, and analog to digital conversion required for each microphone channel. Digital MEMS microphones consolidate these functions to within the microphone element giving a digital output representative of the measured acoustic signal. Digital MEMS microphones allow for a completely digital hardware solution which reduces complexity and allows more flexibility in utilizing higher element number arrays if desired. The performance parameters relevant to HFDSs of current best-in-class digital MEMS microphones were measured and compared to current analog microphones commonly utilized in these systems. The purpose of these tests was to determine the viability of digital MEMS microphones as an alternate acoustic sensing element for a HFDS system being implemented on a multi-rotor drone.

A broadband impulsive signal detection filter for unknown acoustic sources in outdoor environments using non-coplanar microphone arrays

This paper describes a method for non-coplanar microphone arrays that temporally isolates and cleans unknown broadband acoustic impulses for detection, classification, and scene analysis. Possible events are initially identified using a sliding statistical time window. Then the authors posit that most of the false triggers due to environmental noise can be filtered by using generalized cross correlation to phase align the microphone channels and reject implausible velocities. Finally, the phase aligned signals are calibrated and averaged across the microphones. With appropriate hyperparameter tuning, this method appears robust to ambient noise, wind noise and physical interaction. Performance is measured using a simulation and a real historic dataset of over 2 hours of curated acoustic recordings containing 559 gunshots, 120 blasts, and 747 other various weather and non-impulsive events recorded with no prior information under normal operating conditions. Events were found and validated using human listeners with a tool to visualize the waveform and the spectrogram. For this dataset, the model accurately found over 95% of the gunshots with 92% temporal separation and 100% of the blasts identified by the listeners. These results show the method to be a viable solution for impulsive outdoor broadband acoustic signal detection.This paper describes a method for non-coplanar microphone arrays that temporally isolates and cleans unknown broadband acoustic impulses for detection, classification, and scene analysis. Possible events are initially identified using a sliding statistical time window. Then the authors posit that most of the false triggers due to environmental noise can be filtered by using generalized cross correlation to phase align the microphone channels and reject implausible velocities. Finally, the phase aligned signals are calibrated and averaged across the microphones. With appropriate hyperparameter tuning, this method appears robust to ambient noise, wind noise and physical interaction. Performance is measured using a simulation and a real historic dataset of over 2 hours of curated acoustic recordings containing 559 gunshots, 120 blasts, and 747 other various weather and non-impulsive events recorded with no prior information under normal operating conditions. Events were found and validated using human listen...

Arrangements of phased microphone arrays for acoustic source localization based on deconvolution algorithms

Nowadays phased microphone arrays have become a standard technique for acoustic source localization. Until now microphone arrangement is generally based on the conventional beamforming. Compared to the conventional beamforming, deconvolution algorithms such as DAMAS can achieve significantly improved spatial resolution. Recently the computing speeds of deconvolution algorithms have become faster and faster due to the developments of acceleration algorithms and computer ability. In this paper, phased microphone arrays based on deconvolution algorithms is preliminarily investigated in this paper. Three common microphone arrangements (circle array, grid array and spiral array) with same microphone channels are designed firstly. Subsequently their dynamic ranges with deconvolution algorithms are compared. At low frequency, these three arrays have nearly the same dynamic range. However at high frequency, circle array nearly has the same dynamic range with spiral array, while grid array has smallest dynamic range. Considering the experimental costs, circle array could replace spiral array as the first choice of microphone arrangement based on deconvolution algorithms

Development of a multisensor array for localizing bats in space

We present a novel multisensor array to track flight paths of bats in space. Our array is designed to be portable and, thus, field applicable. Eight microphones are employed to localize each echolocation call, whereas two cameras provide continuous information on the flight path between the calls. To reduce costs and, thus, make it available for low-budget research the array was developed from scratch including a novel acoustic signal recording and processing hardware. Our developed acoustic hardware is the first system to sample eight microphone channels synchronously at 1 MS/s/CH with 16 Bit resolution without the necessity of computers or laptops. All eight signals from the ultrasonic microphones are not only recorded but also processed in real time in the frequency as well as time domain to determine the existence of echolocation calls. Only relevant acoustic signal are recorded which enables long term deployments of the array. During the acoustic signal acquisition, the system records images of two infrared cameras at 12 Hz and, in addition, several environmental factors such as temperature, humidity and illumination. Recorded datasets can be associated with a global position and time-stamp through a GPS module and matched to the GPS coordinate system with a three-axis accelerometer. The overall system (microphones, cameras and sensors) was designed to be affordable, easy to use in the field on battery power and without the necessity of an additional laptop. Without cost optimization and only small quantity we reached a prize below 1000 EUR for a complete system.