Chapter Four · failure evidence
What Acoustic & Audio Signal Processing got wrong, from 74 dissertations
The records document engineering challenges and experimental failures across acoustic signal processing, spatial audio rendering, and physical noise control. Investigators repeatedly encountered performance degradations caused by unmodeled environmental acoustics, feature representation deficiencies, model overfitting, and hardware nonlinearities. These records come from PhD theses at 23 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Handcrafted acoustic features and basic thresholding fail to capture complex sound representations
Simple acoustic measures such as baseline MFCCs, time-domain features, and fixed amplitude thresholds frequently lacked discriminative power or sensitivity under high noise conditions. These conventional representations underperformed compared to deep learned embeddings, transcript-derived linguistic features, or domain-specific physical parameters.
Tried and failed
using standard MFCC features alone applied to audio outlier detection. Outcome: worse than baseline. Reason: baseline 13 MFCC features lacked sufficient acoustic representation compared to expanded feature sets
Relational Outlier Detection: Techniques and Applications · Virginia Tech
Tried and failed
discrete wavelet transform and time-domain features applied to acoustic physiological signal classification. Outcome: worse than baseline. Reason: yielded poor specificity compared to spectral mel-frequency cepstral coefficients
Tried and failed
acoustic feature representations for speech classification applied to dementia speech detection. Outcome: worse than baseline. Reason: acoustic representations captured less diagnostic signal than transcript-derived linguistic features
Conversational robots to support well-being and home safety in dementia care · Imperial
Tried and failed
SVM with hand-crafted acoustic features applied to pathological speech detection. Outcome: worse than baseline. Reason: Classical acoustic measures lacked sufficient discriminative power for complex dysarthric speech patterns.
Tried and failed
predicting hand-crafted audio features in auxiliary tasks applied to visual speech recognition front-ends. Outcome: worse than baseline. Reason: hand-crafted spectral features provided weaker representations than self-supervised learned audio representations
Deep audio-visual speech recognition · Imperial
Tried and failed
spectral gating noise reduction applied to audio classification feature extraction. Outcome: worse than baseline. Reason: preprocessing removed subtle acoustic signals relevant to downstream prediction targets
Tried and failed
transverse acoustic sensor classification applied to mechanical degradation detection. Outcome: worse than baseline. Reason: transverse acoustic signals lacked sufficient informative features compared to longitudinal signals, degrading further with smaller sample sizes
Real-time Data Analytics for Condition Monitoring of Complex Industrial Systems · Georgia Tech
Tried and failed
far-field acoustic feature extraction for condition monitoring applied to turbomachinery component degradation. Outcome: worse than baseline. Reason: Acoustic signatures lack sensitivity to component degradation compared to standard thermodynamic parameters.
Investigation of aircraft auxiliary power unit acoustic signatures for condition monitoring · Cranfield
Tried and failed
full-cycle feature extraction without phase gating applied to cyclic acoustic bio-signal classification. Outcome: worse than baseline. Reason: uninformative movement phases diluted signal compared to maneuver-specific informative phases
Analysis of Joint Acoustic Emissions for Health Monitoring with Wearable Technologies · Georgia Tech
Tried and failed
raw unstandardized features for cross-condition classification applied to acoustic anomaly detection. Outcome: did not generalise. Reason: unnormalized feature scale variations degraded cross-condition damage classification performance across operating regimes
Practical and generally applicable condition based maintenance (CBM) system for mud pump · UT Austin
Tried and failed
amplitude-threshold signal detection applied to audio vocalization segmentation. Outcome: worse than baseline. Reason: amplitude thresholding failed to robustly capture syllables compared to deep learning and spectral methods
Lost to a baseline
Traditional machine learning (MFCCs with SVM/classifiers) underperformed deep learning models on non-speech body sounds
Smart Health Technologies Towards Parkinson's Disease Management · DSpace at SUNY Buffalo
Lost to a baseline
Call-level OpenSMILE features (67.95%-70.36% F1) underperformed Call-level x-vectors (72.78%-76.87% F1) and frame-level OpenSMILE (73.63% F1) on acoustic CSAT prediction
Towards Better Understanding of Spoken Conversations: Assessment of Emotion and Sentiment · JScholarship
Considered and rejected
Considered and rejected: Rejected using pitch, loudness, and energy duration features because standard audio processing libraries failed under high background noise and low SNR.
A Machine Learning Based Victim's Scream Detection System for Burning Sites Using an Autonomous Embedded System Vehicle · TXST Digital Repository
Considered and rejected
Considered and rejected: Rejected time-domain waveform representations for audio inference defense because they fail to capture sufficient spatial information compared to Mel-spectrograms.
Robust Defenses Against Adversarial Machine Learning in Internet of Things Security · Carleton University Institutional Repository
Tried and failed
single-threshold crossing time encoding applied to noisy acoustic echo classification. Outcome: did not generalise. Reason: single threshold lacked sufficient dynamic range to capture complex acoustic clutter in real-world environments
Solutions to Passageways Detection in Natural Foliage with Biomimetic Sonar Robot · Virginia Tech
Lost to a baseline
Bag-of-phones feature set using Allosaurus performed poorly compared to aggregated acoustic feature sets across participants
Spatial audio rendering and source localization degrade under acoustic propagation mismatch and simulation limits
Spatial audio systems and localization algorithms suffered from spatial aliasing, head-rotation energy fluctuations, and room reverberation that degraded perceptual sound field quality. Furthermore, models evaluated in laboratory environments or trained on synthetic stimuli failed to match human perception and real-world auditory detection distances.
Lost to a baseline
The 3D 3OA 16-speaker setup yielded worse speech intelligibility in the most reverberant condition compared to a 2D Vector Base Amplitude Panning setup with higher frontal speaker density.
Set-up and investigation of a virtual reality system for clinical usage aimed at more ecological hearing assessments · IRIS - POLITO - prod
Considered and rejected
Considered and rejected: Single-track F0 tracking was rejected for overlapping speech because single-pitch estimators fail when multiple speakers talk simultaneously.
Tried and failed
magnitude least squares with covariance constraint applied to binaural spatial audio rendering. Outcome: worse than baseline. Reason: did not outperform standard magnitude least squares in perceptual spectral difference or externalisation metrics
Improving binaural audio techniques for augmented reality · Imperial
Tried and failed
dynamic low-order spatial audio rendering applied to binaural room impulse response synthesis. Outcome: worse than baseline. Reason: loudness instability and energy fluctuations during dynamic head rotation
Improving binaural audio techniques for augmented reality · Imperial
Tried and failed
distribution normality testing for perceptual model selection applied to spatial audio transfer function selection. Outcome: no signal. Reason: metric failed to differentiate between best, worst, and individual transfer functions in localization performance
Auditory model-based similarity metric for head-related transfer functions · Imperial
Tried and failed
deep neural source separation networks applied to psychoacoustic auditory scene analysis illusions. Outcome: did not generalise. Reason: networks trained on audio failed to replicate human perceptual grouping and scene decomposition judgments
Tried and failed
ambisonic loudspeaker array playback applied to acoustic source localization algorithms. Outcome: worse than baseline. Reason: Spatial aliasing and multiple coherent loudspeakers degraded high-resolution direction-of-arrival estimation accuracy.
On the spatial and resilience-related features of natural soundscapes · Imperial
Tried and failed
perceptual audio quality objective metric applied to audio distortion and artifact assessment. Outcome: did not generalise. Reason: Inconsistent predictions for broadband noise, bandwidth reduction, and neural artifacts.
Towards perceptual metrics for audio quality assessment · Georgia Tech
Tried and failed
sensor directivity simulation for domain adaptation applied to spatial audio perception models. Outcome: worse than baseline. Reason: simulated directional sensor patterns degraded transfer to real directional hardware compared to simpler omnidirectional models
Deep Learning Approaches for Auditory Perception in Robotics · EPFL
Tried and failed
multichannel loudspeaker virtual acoustic reproduction applied to acoustic pedestrian detection distance evaluation. Outcome: did not generalise. Reason: laboratory spatial sound field failed to replicate real-world auditory detection distances, underestimating on-road performance
An Evaluation of Laboratory and Test Road Environments and Electric Vehicle Warning Sounds and Systems · Virginia Tech
Tried and failed
perceptual model localization error normality testing applied to spatial audio filter selection. Outcome: no signal. Reason: frontal-only error distribution normality did not correlate above chance with subjective trajectory-based human perception ratings
Auditory model-based similarity metric for head-related transfer functions · Imperial
Tried and failed
Acoustic beamforming source localization applied to Coupled high-frequency aerodynamic resonators. Reason: Out-of-phase inter-resonator coupling disrupted spatial localization
Novel ultrasonic bat deterrents based on aerodynamic whistles · Iowa State
Tried and failed
differentiable fluid wave propagation inverse design applied to acoustic hologram phase optimization. Outcome: did not generalise. Reason: neglected shear mode conversion and multiple internal reflections in solid media for complex high-spatial-frequency targets
Transcranial FUS Therapy and Monitoring using Nonlinear Acoustics · Georgia Tech
Lost to a baseline
Few-Shot RIR baseline underperformed significantly on ACOUSTICROOMS due to UNet reconstruction bottlenecks and mismatched binaural echo representations
Towards Multi-Modal Interactive Systems that Connect Audio, Vision and Beyond · ResearchWorks
Considered and rejected
Considered and rejected: Rejected static-head-only sound localization testing, because it underestimates real-world acoustic performance.
Active listening in sound localization: Multisensory and motor contributions to perceiving and re-learning the auditory space · IRIS - UNITN - prod
Considered and rejected
Considered and rejected: Virtual audio presentation over earphones was evaluated and abandoned because it failed to produce compelling externalization and lacked real head-worn device relevance
On the plasticity of sound localization during the use of hearing protection devices · ResearchWorks
Neural models overfit training distributions and fail to generalize across recording environments
Deep neural architectures and spike-encoding schemes frequently memorized training data artifacts, overfit when trained exclusively on synthetic audio, or failed when model complexity exceeded dataset size. Consequently, classification performance collapsed when tested across new physical recording sites, unseen subjects, or acoustically altered signals.
Tried and failed
auditory-inspired feature extraction with recurrent networks applied to speech recognition under severe amplitude distortion. Outcome: did not generalise. Reason: features failed to retain sufficient acoustic information under extreme peak clipping compared to human perception
Features of hearing: applications of machine learning to uncover the building blocks of hearing · Imperial
Tried and failed
training neural network on synthetic unnatural stimuli applied to spatial audio localization models. Outcome: did not generalise. Reason: models failed to learn fine elevation spectral cues without natural acoustic statistics
Modeling and Evaluating Human Sound Localization in the Natural Environment · MIT
Tried and failed
diffusion-based image inpainting data augmentation applied to spectrogram bioacoustic signal detection. Outcome: did not generalise. Reason: precision dropped substantially when evaluated across different geographic recording environments
Passive Acoustic Monitoring of Whales and Ocean Ambient Sound to Inform Management · Cornell
Tried and failed
neuromorphic spike encoding with auditory filterbanks applied to acoustic echo classification. Outcome: overfit. Reason: high-dimensional spike matrices caused model overtraining on large acoustic datasets
Biomimetic Detection of Dynamic Signatures in Foliage Echoes · Virginia Tech
Tried and failed
neuro-inspired feature extraction with recurrent networks applied to speech recognition under acoustic degradation. Outcome: did not generalise. Reason: models degraded on noise-vocoded speech but tolerated periodic speech, contradicting human perceptual robustness
Features of hearing: applications of machine learning to uncover the building blocks of hearing · Imperial
Tried and failed
increasing model depth on pretrained embeddings applied to audio event classification. Outcome: did not generalise. Reason: deeper models fit training distribution artifacts but degraded on real-world ambient sounds
Towards practical acoustic sensing for human activity recognition on commercial wearable devices · UT Austin
Tried and failed
fine-tuning models solely on synthetic generated data applied to automatic speech recognition models. Outcome: overfit. Reason: training exclusively on synthetic data led to overfitting and degraded generalization to genuine speech
Voice conversion and text-to-speech for privacy protection applications · Imperial
Tried and failed
Lowering classification head dropout during fine-tuning applied to audio spectrogram transformer classification. Outcome: overfit. Reason: Insufficient regularization led to higher training accuracy at the expense of test generalization
AI-generated synthetic audio analysis and forensic attribution · Iowa State
Tried and failed
overlapping window segmentation with weak regularization applied to audio regression from spectrogram representations. Outcome: overfit. Reason: overlapping windows caused session-level memorization under MSE loss, severely degrading generalization
Tried and failed
Machine learning on audio features with confounder matching applied to respiratory acoustic disease detection. Outcome: did not generalise. Reason: Severe dataset shift and acoustic recording confounders across independent collection environments.
Unsound foundations: refining AI’s role in audio-based COVID-19 detection · Imperial
Tried and failed
deep learning on pseudonymized audio signals applied to pathological speech classification. Outcome: did not generalise. Reason: pseudonymization distorted acoustic cues essential for distinguishing pathology from healthy controls
Novel Methods for Incorporating Prior Knowledge for Automatic Speech Assessment · EPFL
Tried and failed
handcrafted spatiotemporal and spectral feature extraction applied to acoustic emission classification across unseen subjects. Outcome: did not generalise. Reason: Handcrafted features failed to capture subject-invariant patterns across unseen individuals compared to learned representations
Quantifying the Effects of Knee Joint Biomechanics on Acoustical Emissions · Georgia Tech
Tried and failed
Neuromorphic spike encoding of acoustic echo signatures applied to Outdoor vegetation acoustic classification. Outcome: did not generalise. Reason: Spatial variance across physical recording locations degraded classification performance to chance level.
Biomimetic Detection of Dynamic Signatures in Foliage Echoes · Virginia Tech
Tried and failed
1D convolutional neural network on spectrograms applied to audio classification from small dataset. Outcome: overfit. Reason: Excessive model parameter complexity relative to the training dataset size
AI for Dementia Detection: Open AI Health Diagnostic (OAHD) · Harvard
Lost to a baseline
On real-world data with log-STFT (window 128), Xception+BAM achieved an F1-score of only 55.85% compared to AcousticNet's 83.18%.
Advanced Deep Learning Models for Acoustic Event Detection in Prestressed Concrete Bridge Monitoring · IRIS - POLITO - prod
Physical and active noise reduction techniques fail under airborne paths, flow saturation, and urban reverberation
Active noise control and impedance actuators saturated under high flow noise, exhibited microdischarge nonlinearities, or diverged when confronted with temporal clearance changes. Passive damping patches, barriers, and porous blankets failed because airborne noise, low-frequency tones, and reverberation bypassed structural attenuation or increased total radiated sound power.
Tried and failed
dual-microphone acoustic noise cancellation applied to low-frequency motor interference removal. Reason: failed to eliminate narrow-band interference around 110 Hz
Tried and failed
active impedance control using electroacoustic actuators applied to acoustic absorption in flow ducts. Reason: flow noise exceeded operational limits, saturating the actuator control voltage
Plasma-based Electroacoustic Actuator for Broadband Sound Absorption · EPFL
Tried and failed
passive acoustic attenuation blankets applied to industrial chiller low-frequency noise. Reason: passive porous blankets poorly attenuate tonal low-frequency acoustic emissions
Cloud Ecologies: An Environmental Ethnography of Data Centers · MIT
Tried and failed
dielectric barrier discharge actuators applied to active acoustic impedance control. Reason: pulse microdischarges caused severe acoustic nonlinearity under sinusoidal drive
Plasma-based Electroacoustic Actuator for Broadband Sound Absorption · EPFL
Tried and failed
Acoustic sensor measurement at receiver locations applied to isolating specific urban noise sources. Outcome: no signal. Reason: Ambient background noise consistently exceeded regulatory thresholds, obscuring target source contributions
Noise, Sound, And Urban Coexistence: Interrogating Practices Of Sonic Management In Mexico City · Penn
Tried and failed
constrained layer damping patch placement applied to assembled appliance acoustic noise reduction. Outcome: no signal. Reason: Airborne motor and exhaust noise overwhelmed the structural shell vibration noise reductions.
Using selectively placed patches of constrained layer damping to quiet home appliances · Iowa State
Tried and failed
protrusive acoustic screening devices with sound absorbers applied to building facade noise reduction. Outcome: worse than baseline. Reason: urban reverberation undermines the directional sound screening efficacy even with absorption material
Tried and failed
constant gain PID feedback control applied to time-varying acoustic emission regulation. Outcome: unstable. Reason: spatial density variations and temporal clearance kinetics caused control loop divergence
Microbubble Dynamics Monitoring and Control for Diagnosis and Treatment of Brain Cancer · Georgia Tech
Tried and failed
dynamic sound masking based on ambient noise applied to multi-speaker speech perception. Outcome: worse than baseline. Reason: added acoustic energy increased cognitive processing load instead of improving intelligibility
Designing Responsive Environments to Support Speech Perception for Individuals with Mild Cognitive Impairment · Georgia Tech
Tried and failed
fine-grained environmental noise categorization using mixture of experts applied to robust speech recognition front-end. Outcome: worse than baseline. Reason: specialized multi-class noise conditioning offered no benefit over binary clean versus noisy distinction
Modelling cochlea and its interaction with the auditory path for speech processing · EPFL
Tried and failed
empirical acoustic regression against meteorological variables applied to environmental noise propagation modeling. Reason: empirical correlation contradicted known physical acoustic attenuation laws due to unmodeled operational confounders
A Data-Driven Approach to Departure and Arrival Noise Abatement Flight Procedure Development · MIT
Tried and failed
selective constrained layer damping patch placement applied to appliance shell structural noise reduction. Outcome: worse than baseline. Reason: modified acoustic radiation coupling and disrupted destructive interference between vibrating modes, increasing radiated sound power
Using selectively placed patches of constrained layer damping to quiet home appliances · Iowa State
Lost to a baseline
Acoustic side-channel monitoring exhibited worse noise tolerance / higher susceptibility to noise-induced errors than EM monitoring below 25 dB SNR.
Securing Cyber-Physical Systems by Improving and Optimizing Measurement of the Electromagnetic Backscattering Side-Channel · Georgia Tech
Considered and rejected
Considered and rejected: PDMS-encapsulated transducers caused slight acoustic attenuation and energy dissipation due to viscoelastic absorption, requiring compensation via higher input voltages to maintain acoustic radiation force.
Acoustic Assembly of Collagen Hydrogels in Petri Dishes: The Distinct Roles of Traveling and Standing Waves · Virginia Tech
Considered and rejected
Considered and rejected: Rejected measuring acoustic noise thrusting away from the stand because wake impingement on the stand and nearby building corrupted -90 deg AoE microphone data
A comprehensive study of a coaxial, co-rotating rotor in hover · UT Austin
Speech enhancement and echo filtering produce spectral distortion and target signal cancellation
Techniques such as spectral peak amplification, multichannel Wiener filtering, and inverse filtering introduced severe phonetic distortion, spectral tilt, or target speech cancellation. In addition, optimization objectives tied purely to signal-to-noise ratio or symmetric echo cancellation assumptions degraded intelligibility cues and failed under acoustic channel estimation errors.
Tried and failed
selective spectral peak amplification applied to speech signal enhancement for human perception. Outcome: worse than baseline. Reason: amplifying second formant peaks distorted phonetic cues, degrading recognition of lax vowels and velar stops
Tried and failed
selective spectral peak amplification applied to speech enhancement in noisy conditions. Outcome: worse than baseline. Reason: linear amplification introduced spectral tilt distortion that degraded vowel identification in high noise
Tried and failed
voice activity detection based multichannel wiener filtering applied to speech enhancement at positive SNRs. Outcome: worse than baseline. Reason: speech leakage into noise covariance estimate caused target signal cancellation
Speech enhancement in microphone array networks using polynomial matrix decomposition · Imperial
Tried and failed
optimizing complex neural networks solely for SNR applied to binaural speech enhancement under reverberant conditions. Outcome: did not generalise. Reason: optimizing SNR alone degraded binaural speech intelligibility cues despite high SNR gains
Binaural enhancement of speech for hearing assistive devices · Imperial
Considered and rejected
Considered and rejected: Rejected using the Multi-channel Inversion Theorem (MINT) for multi-channel speech dereverberation due to its lack of robustness to acoustic channel estimation errors.
Considered and rejected
Considered and rejected: Rejected using anechoic speech signal as recovery target in favor of the direct path signal because it failed to account for propagation delay, misidentifying phoneme onset/termination artifacts.
Mitigating Cochlear Implant Stimulus Pulses Dominated by Reverberant Distortions · DukeSpace
Tried and failed
deep pretrained perceptual feature losses applied to speech denoising and audio enhancement. Outcome: worse than baseline. Reason: deep feature representations did not outperform simple biologically-inspired filter bank loss functions
Modeling and Evaluating Human Sound Localization in the Natural Environment · MIT
Tried and failed
thresholding time-domain angle deviation for echo rejection applied to broadband acoustic split-beam angle estimation. Outcome: did not generalise. Reason: introduces false rejection bias against broadband single echoes exhibiting resonant scattering phase distortions
Broadband Acoustic Characterization of the Deep Scattering Layer · MIT
Tried and failed
comb liftering in the complex cepstrum domain applied to acoustic multipath echo cancellation. Reason: distorted echoes caused energy to leak across adjacent quefrency bins, making discrete notch liftering ineffective
Application of the cepstrum technique to location of acoustic sources in the presence of a reflective surface · Virginia Tech
Tried and failed
nonlinear oscillator filterbank front-end applied to speech recognition under acoustic noise. Outcome: worse than baseline. Reason: did not provide superior noise robustness compared to conventional convolutional filters in clean-to-noise transfer
Modelling cochlea and its interaction with the auditory path for speech processing · EPFL
Lost to a baseline
When there is only far-end speech without near-end disturbance, the adaptive filter alone outperforms the correlation method with higher ERLE and faster tracking.
Correlation-based acoustic echo canceler Correlation based acoustic echo canceler · Iowa State
Considered and rejected
Considered and rejected: Rejected acoustic echo cancellation (AEC) for concurrent music separation because it assumes symmetric spectral distortions and only works for simple Doppler presence detection
Fine-grained wireless perception for human tracking · Cornell
Considered and rejected
Considered and rejected: Rejected equal error rate (EER) thresholding typical of standard VADs because speech enhancement has asymmetric sensitivity (low tolerance for speech classified as noise, high tolerance for noise classified as speech)
Voice inactivity ranking for enhancement of speech on microphone arrays · OpenBU
Considered and rejected
Considered and rejected: Rejected dynamic time-warping due to high sensitivity to differences in background noise and reduced accuracy with lossy compressed audio formats (mp3).
Acoustic Signals of Closely Related Bird Species Differ More in Sympatry Than in Allopatry - But Not If They Are Learned · Queens University Institutional Repository
Left open by the authors
Problems the authors named and did not get to.
Left open
Combine phase-related loss formulations supporting multiple valid solutions with the consistency-preserving loss LEC for speech enhancement models. Blocker: None
Incorporating Geometric and Consistency Constraints into Deep Models for Robust Phase Reconstruction and Speech Enhancement · Georgia Tech
Left open
Evaluate whether machine learning model embeddings of beamformed audio signals form sub-clusters within omnidirectional soundscape embedding spaces. Blocker: Requires multichannel soundscape recording datasets or the proprietary MAARU hardware array recordings
On the spatial and resilience-related features of natural soundscapes · Imperial
Left open
Evaluate varying F2 formant enhancement gain levels across speech datasets to balance objective intelligibility metrics against spectral distortion. Blocker: Requires conducting human perceptual listening tests or using objective speech intelligibility and distortion models.
Left open
Evaluate speech waveform codec SNR performance using test signals composed of 3 or 4 combined sinusoids. Blocker: None
PERFORMANCE ANALYSIS OF SPEECH WAVEFORM CODECS USING A DUAL SINUSOID INPUT · HARVEST
Left open
Develop speech enhancement models robust to cross-device and cross-microphone variations in bone-conducted speech signals. Blocker: None
Prediction and Enhancement of Speech Intelligibility in Challenging Acoustic Environments · Queens University Institutional Repository
Left open
Analyze onset responses in continuous speech sEEG/EEG data during speech errors and non-canonical utterances using receptive field models. Blocker: Requires invasive human sEEG/EEG recordings during continuous speech production tasks containing speech errors
Cortical suppression of auditory feedback during speech production and perception · UT Austin
Left open
Extend the indirect noise-aware speech enhancement framework to multi-channel inputs, complex ratio masking, and raw waveform architectures. Blocker: None
An Indirect Speech Enhancement Framework Through Intermediate Noisy Speech Targets · Georgia Tech
Left open
Implement adaptive and MVDR beamforming integrated with monophonic source separation on the autonomous acoustic recording hardware. Blocker: Requires the physical Multichannel Acoustic Autonomous Recording Unit (MAARU) hardware or specialized multichannel microphone array apparatus.
On the spatial and resilience-related features of natural soundscapes · Imperial
Left open
Analyze microphone-recorded acoustic waveform signatures to detect distribution transformer degradation and partial discharge events. Blocker: Requires acoustic sensor data from physical transformers experiencing degradation or partial discharge.
DISTRIBUTION TRANSFORMER ASSET MONITORING ON THE GRID EDGE USING SMART SENSOR DATA · Georgia Tech
Left open
Design and evaluate reduced audio feature vectors using salient features like Chroma and CENS for efficient Ragam classification model training. Blocker: None
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.