Causal inference in environmental sound recognition.

Sound is caused by physical events in the world. Do humans infer these causes when recognizing sound sources? We tested whether the recognition of common environmental sounds depends on the inference of a basic physical variable - the source intensity (i.e., the power that produces a sound). A source's intensity can be inferred from the intensity it produces at the ear and its distance, which is normally conveyed by reverberation. Listeners could thus use intensity at the ear and reverberation to constrain recognition by inferring the underlying source intensity. Alternatively, listeners might separate these acoustic cues from their representation of a sound's identity in the interest of invariant recognition. We compared these two hypotheses by measuring recognition accuracy for sounds with typically low or high source intensity (e.g., pepper grinders vs. trucks) that were presented across a range of intensities at the ear or with reverberation cues to distance. The recognition of low-intensity sources (e.g., pepper grinders) was impaired by high presentation intensities or reverberation that conveyed distance, either of which imply high source intensity. Neither effect occurred for high-intensity sources. The results suggest that listeners implicitly use the intensity at the ear along with distance cues to infer a source's power and constrain its identity. The recognition of real-world sounds thus appears to depend upon the inference of their physical generative parameters, even generative parameters whose cues might otherwise be separated from the representation of a sound's identity.

Machine learning in acoustics: Theory and applications.

Acoustic data provide scientific and engineering insights in fields ranging from biology and communications to ocean and Earth science. We survey the recent advances and transformative potential of machine learning (ML), including deep learning, in the field of acoustics. ML is a broad family of techniques, which are often based in statistics, for automatically detecting and utilizing patterns in data. Relative to conventional acoustics and signal processing, ML is data-driven. Given sufficient training data, ML can discover complex relationships between features and desired labels or actions, or between features themselves. With large volumes of training data, ML can discover models describing complex acoustic phenomena such as human speech and reverberation. ML in acoustics is rapidly developing with compelling results and significant future promise. We first introduce ML, then highlight ML developments in four acoustics research areas: source localization in speech processing, source localization in ocean acoustics, bioacoustics, and environmental sounds in everyday scenes.

Headphone screening to facilitate web-based auditory experiments.

Psychophysical experiments conducted remotely over the internet permit data collection from large numbers of participants but sacrifice control over sound presentation and therefore are not widely employed in hearing research. To help standardize online sound presentation, we introduce a brief psychophysical test for determining whether online experiment participants are wearing headphones. Listeners judge which of three pure tones is quietest, with one of the tones presented 180° out of phase across the stereo channels. This task is intended to be easy over headphones but difficult over loudspeakers due to phase-cancellation. We validated the test in the lab by testing listeners known to be wearing headphones or listening over loudspeakers. The screening test was effective and efficient, discriminating between the two modes of listening with a small number of trials. When run online, a bimodal distribution of scores was obtained, suggesting that some participants performed the task over loudspeakers despite instructions to use headphones. The ability to detect and screen out these participants mitigates concerns over sound quality for online experiments, a first step toward opening auditory perceptual research to the possibilities afforded by crowdsourcing.

Statistics of natural reverberation enable perceptual separation of sound and space.

In everyday listening, sound reaches our ears directly from a source as well as indirectly via reflections known as reverberation. Reverberation profoundly distorts the sound from a source, yet humans can both identify sound sources and distinguish environments from the resulting sound, via mechanisms that remain unclear. The core computational challenge is that the acoustic signatures of the source and environment are combined in a single signal received by the ear. Here we ask whether our recognition of sound sources and spaces reflects an ability to separate their effects and whether any such separation is enabled by statistical regularities of real-world reverberation. To first determine whether such statistical regularities exist, we measured impulse responses (IRs) of 271 spaces sampled from the distribution encountered by humans during daily life. The sampled spaces were diverse, but their IRs were tightly constrained, exhibiting exponential decay at frequency-dependent rates: Mid frequencies reverberated longest whereas higher and lower frequencies decayed more rapidly, presumably due to absorptive properties of materials and air. To test whether humans leverage these regularities, we manipulated IR decay characteristics in simulated reverberant audio. Listeners could discriminate sound sources and environments from these signals, but their abilities degraded when reverberation characteristics deviated from those of real-world environments. Subjectively, atypical IRs were mistaken for sound sources. The results suggest the brain separates sound into contributions from the source and the environment, constrained by a prior on natural reverberation. This separation process may contribute to robust recognition while providing information about spaces around us.

Avoiding a lost generation of scientists.

By sharing their experiences, early-career scientists can help to make the case for increased government funding for researchers.

Compressive geoacoustic inversion using ambient noise.

Surface generated ambient noise can be used to infer sediment properties. Here, a passive geoacoustic inversion method that uses noise recorded by a drifting vertical array is adopted. The array is steered using beamforming to compute the noise arriving at the array from various directions. This information is used in two different ways: Coherently (cross-correlation of upward/downward propagating noise using a minimum variance distortionless response fathometer), and incoherently (bottom loss vs frequency and angle using a conventional beamformer) to obtain the bottom properties. Compressive sensing is used to invert for the number of sediment layer interfaces and their depths using coherent passive fathometry. Then the incoherent bottom loss estimate is used to refine the sediment thickness, sound speed, density, and attenuation values. Compressive sensing fathometry enables automatic determination of the number of interfaces. It also tightens the sediment thickness priors for the incoherent bottom loss inversion which reduces the search space. The method is demonstrated on drifting array data collected during the Boundary 2003 experiment.

Ocean bottom profiling with ambient noise: a model for the passive fathometer.

A model is presented for the complete passive fathometer response to ocean surface noise, interfering discrete noise sources, and locally uncorrelated noise in an ideal waveguide. The leading order term of the ocean surface noise contribution produces the cross-correlation of vertical multipaths and yields the depth of sub-bottom reflectors. Discrete noise incident on the array via multipaths give multiple peaks in the fathometer response. These peaks may obscure the sub-bottom reflections but can be attenuated with use of minimum variance distortionless response (MVDR) steering vectors. The seabed critical angle introduces discontinuities in the spatial distribution of distant surface noise and may introduce spurious peaks in the passive fathometer response. These peaks can be attenuated by beamforming within a bandwidth limited by the array geometry and critical angle.

Coherent averaging of the passive fathometer response using short correlation time.

The passive fathometer algorithm was applied to data from two drifting array experiments in the Mediterranean, Boundary 2003 and 2004. The passive fathometer response was computed with correlation times from 0.34 to 90 s and, for correlation times less than a few seconds, the observed signal-to-noise ratio (SNR) agrees with a 1D model of SNR of the passive fathometer response in an ideal waveguide. In the 2004 experiment, the fathometer response showed the array depth varied periodically with an amplitude of 1 m and a period of 7 s consistent with wave driven motion of the array. This introduced a destructive interference, which prevents the SNR growing with increasing correlation time. A peak-tracking algorithm applied to the fathometer response of experimental data was used to remove this motion allowing the coherent passive fathometer response to be averaged over several minutes without destructive interference. Multirate adaptive beamforming, using 90 s correlation time to form adaptive steer vectors which were applied to 0.34 s data snapshots, increases the SNR of the passive fathometer response.

On the sign of the adaptive passive fathometer impulse response.

Harrison [J. Acoust. Soc. Am. 125, 3511-3513 (2009)] presented a mathematical explanation for a sign-inversion induced to the passive fathometer response by minimum variance distortionless response (MVDR) beamforming. Here a concise mathematical formulation is offered, which decomposes the cross-spectral density matrix into coherent and incoherent components and allows the matrix inversion to be obtained exactly by eigendecomposition. This shows that, in the region containing the bottom reflection, the MVDR fathometer response is identical to that obtained with conventional processing multiplied by a negative factor.

Shallow-water seismoacoustic noise generated by tropical storms Ernesto and Florence.

Land-based seismic observations of double frequency (DF) microseisms generated during tropical storms Ernesto and Florence are dominated by signals in the 0.15-0.5 Hz band. In contrast, data from sea floor hydrophones in shallow water (70 m depth, 130 km off the New Jersey coast) show dominant signals in the ocean gravity-wave frequency band, 0.02-0.18 Hz, and low amplitudes from 0.18 to 0.3 Hz, suggesting significant opposing wave components necessary for DF microseism generation were negligible at the site. Florence produced large waves over deep water while Ernesto only generated waves in coastal regions, yet both storms produced similar spectra. This suggests near-coastal shallow water as the dominant region for observed microseism generation.