Research of Daniele Salvati

Audio and Acoustic Signal Processing

Audio and acoustic signal processing involves the study, modeling, and manipulation of audio and acoustic signals. The field includes recording and playback, data compression, filtering, signal enhancement, and recognition. Key topics in audio and acoustic signal processing include:

• Audio modeling, coding, and transmission
• Spatial audio recording and reproduction
• Signal enhancement
• Source separation
• System identification and reverberation reduction
• Echo reduction
• Acoustic sensor array processing
• Auditory modeling
• Detection and classification of acoustic scenes and events
• Analysis and synthesis of acoustic environments
• Analysis, processing, and synthesis of musical signals
• Music information retrieval
• Bioacoustics and medical acoustics
• Audio security

Audio Signal Processing in the 21st Century

Computer Audition

Computer audition is a subfield of audio and acoustic signal processing focused on the automatic analysis, understanding, and interpretation of auditory information. This interdisciplinary area combines signal processing, machine learning, and psychoacoustics. Applications of computer audition include acoustic scene analysis, music analysis, human-computer interaction, and surveillance.

Acoustic Scene Analysis

Acoustic scene analysis involves the automatic analysis and understanding of acoustic environments. The main tasks of an acoustic scene analysis system include acoustic event detection, which determines when an event occurs; acoustic source localization, which estimates where it occurs; acoustic event classification, which identifies what or who produced the sound; and acoustic scene classification.

Acoustic Source Localization

The goal of an acoustic source localization system is to estimate the position of acoustic sources in space by analyzing the sound field captured by a microphone array, which is a set of microphones arranged to capture spatial information about sound. Spatial localization of acoustic sources has received considerable attention, and baseline techniques are now available that provide effective performance in a wide range of real-world conditions, including indoor and outdoor scenarios, reverberant and noisy environments, and near-field and far-field monitoring.

Today, acoustic source localization methods can be broadly classified into two categories: TDOA-based indirect methods and direct methods. Indirect methods aim to estimate the time differences of arrival of the acoustic wavefront between microphone pairs and then infer the source position through geometric considerations. The generalized cross-correlation (GCC) is considered a practical baseline method for TDOA estimation, although improved versions are often used in practice. The multichannel cross-correlation coefficient (MCCC), for example, is based on TDOA estimates obtained through GCC and combined with a prediction of spatial error to provide a more robust estimate of the source position. Direct methods, on the other hand, estimate the position of an acoustic source in a single step by exploiting a power density function representing the distribution of spatially relevant information, and they are generally considered more robust than TDOA-based methods in noisy and reverberant conditions. SRP localization involves computing the output power of a beamformer steered toward each target position of interest. Conventional SRP is performed with the delay-and-sum beamformer, which consists of synchronizing the array signals to steer the array in a given direction and summing the signals to estimate the power of the spatial filter. The SRP phase transform (SRP-PHAT) is a widely used filtered SRP beamforming method. The PHAT filter assigns equal importance to each frequency by normalizing the spectrum by its magnitude. SRP-PHAT can be efficiently computed through the global coherent field (GCF) approach, which coherently sums the GCC-PHAT functions from the microphone pairs for each possible point of interest. Among conventional beamformers, the minimum variance distortionless response (MVDR) filter is a well-known data-dependent beamformer that provides higher resolution than conventional beamforming. Another class of high-resolution methods is based on subspace analysis and decomposition. The multiple signal classification (MUSIC) method exploits the subspace orthogonality property to build the spatial spectrum and localize sources in terms of direction of arrival. The estimation of signal parameters via rotational invariance techniques (ESPRIT) is also based on subspace decomposition and exploits rotational invariance. The diagonal unloading beamforming (DUB) provides high resolution with low computational complexity by subtracting a suitable diagonal matrix from the covariance matrix, thereby removing as much of the signal subspace as possible from the covariance matrix.