Stephen Roessner

Audio phase singularity detection for room acoustics parameter estimation

An acoustical measurement method is presented for room acoustical parameter estimation. The proposed method collects the response of sinusoidal test signals and detects the audio phase singularity points as room acoustical features.

Audio frequency response of glass flat panel distributed mode loudspeakers

The development of glass flat panel distributed mode loudspeakers (DML’s) will enable compelling new applications, such as flat panel displays that double as loudspeakers, or in architectural and automotive applications, window glass that can produce sound or actively cancel environmental noise. In this paper we discuss the frequency response of glass flat-panel distributed mode loudspeakers. The density of panel bending modes, their quality factors (Q’s), and the modal radiation efficiencies determine a DML’s frequency response. For example, a 0.55 mm thick cover glass for a 55 in. television has more than 10 000 bending modes in the audible frequency range. The density of modes increases with frequency and above a threshold frequency, determined by the panel dimensions and the mode Q’s, the mode spacing becomes less than the width of the individual modes, so the overall frequency response approaches the smooth, flat response of an ideal pistonic loudspeaker. However, below this frequency, discrete modes produce prominent peaks in the frequency response. We present simulated and measured glass panel frequency responses, including the effect of a thin layer of trapped air behind a DML panel, and we discuss the need for high internal friction (low Q) glass.

Transient response of flat-panel distributed mode loudspeakers

The acoustic impulse response of a loudspeaker must have short rise and decay times to provide accurate reproduction of signals with fast transients, such as human speech. In a flat panel distributed mode loudspeaker (DML), the transverse bending waves that are the primarily source of acoustic radiation, propagate across the panel at finite velocity. Therefore, regions of the panel more distant from the panel driving point(s) radiate sound at later times, which broadens the DML’s acoustic impulse response. Furthermore, bending wave propagation in a rigid panel is dispersive, with a wave velocity that is proportional to the square root of the frequency, which creates additional distortions of signals with fast transients. Mechanical simulations and scanning laser vibrometer measurements of glass panel DML’s clearly illustrate the delayed radiation and dispersion distortion mechanisms. Furthermore, measurements of the acoustic impulse response of DML’s are in agreement with simulations, which provides addit...

An additional “Depth” of reverberation helps content stand out: Media content emphasis using audio reverberation effect

The audio reverberation effect adds in a perceptual depth dimension to media content and can be intelligently applied to emphasis the target media segments. A quantitative model of this emphasis effects is derived from subjective evaluation experiments.

Interacting with semantic musical features: Enhanced human-computer interactions based on musical “meaning”

Semantic musical features are proposed to reflect the understanding of the music, instead of the music itself, and serve as idea interfaces for musical “meaning” based human-computer interactions. The proposed semantic musical features are based on reductive music analysis and musical expressive features.

Measurement and analysis of timing asynchronies in ensemble performance

Timing asynchrony is an important timing descriptor of ensemble music performance. In this paper the timing asynchrony is measured as offsets between “concurring” music onsets. Specifically we measure the offset between music note onsets that are prescribed to be synchronized according to the music score. First, we conduct measurements on the multi-track audio with separate instrument tracks. These multi-track materials are recorded by using acoustically isolated recording booth or by conducting multiple recording rounds. We then perform statistical analysis both on individual asynchrony points and on asynchrony points at key coordination locations, such as the start and the end of a music phrase. We emphasize in the analysis that the musical timing asynchronies should not be treated only as performance discrepancies because part of these micro-deviation patterns provide artistic “lively” elements. We also generalize the proposed framework to mixed-down polyphonic recordings. For polyphonic recordings whe...

Randomized sample-level interpolation for audio content manipulation

In this paper we propose a novel audio signal manipulation algorithm based on sample-level interpolations that can generate multiple unique versions of an audio file without creating any perceptual difference. The proposed algorithm enables important applications such as digital rights management and file distribution tracking. The simplest sample-level interpolation method is based on time domain interpolation of fixed-length audio frames. The processing algorithm first segments the audio signal into fixed-length frames. For each frame, we perform a slight time compression or extension using an audio sample interpolation algorithm and then we recombine the manipulated audio samples to form a manipulated version of the original audio files. To enable better security features a randomization program is applied to control the frame-length and manipulation-length using pseudo-random sequences. The result of this algorithm is effectively a form of weak frequency modulation. If the fame size is larger than the compression/extension sample number, these compression/extension manipulations will not produce any audible difference. Various subjective evaluation experiments are conducted to decide the extent of the admissible processing parameters that will not cause noticeable difference in both fixed-length and randomized-length sample manipulation. The authors also provide several implementation examples and a brief summary.

Real-time audio signal capture and processing using MATLAB object oriented programming

In MATLAB programming language the real-time audio processing functions are usually simulated in non-real-time due to a lack of real-time audio programming support. As a result the real-time audio signal capture and processing functionalities are usually implemented in other programming languages and cannot utilize the extensive signal processing functionalities provided by MATLAB. In this paper we introduce a MATLAB real-time signal processing framework based on MATLAB timer object and audiorecorder object. The proposed processing framework serves as an alternative solution for real-time programming implementation and demonstration. In our proposed processing framework the timer object is implemented to handle the looping of processing cycle, schedule the signal processing tasks, and handle the error processing. The audio capturing/processing functionality is implemented in the timer cycle by using two audiorecorder objects that read the audio streaming data and feed a segment of data to signal processing alternatively. The proposed framework achieves satisfactory real-time performance with no missing audio frames when a short audio delay setting of 10ms is applied. Several application examples of our proposed framework are also demonstrated.

Extracting heterogeneous structured features from music performance

In music performance, the musician adds artistic expressive elements beyond the information contained in conventional western music scores. As micro-fluctuations these expressive dimensions convey emotions and essential interpretative information and can be measured and compared quantitatively, over large and small scales, and evaluated for their effect on aspects of the performance. We present a heterogeneous expressive music feature description that include both inter-note features that extend over musical phrases composed of several notes, and intra-note features that represent the internal variations within each musical note. The intra-note features include pitch and pitch deviation, dynamic level, timber, articulation and vibrato. The inter-note features include timing and dynamics, as well as timber, pitch deviation, articulation, and vibrato extending across multiple notes and musical phrases. A complete multi-dimensional feature description for every note is unnecessary because there is a hierarch...

Temporal modeling of measurement conditions to enhance room acoustical parameter characterization

Forming statistical combinations of the results of repeated acoustical measurements taken under identical conditions is a common practice to reduce the effects of random noise. The most common method is to calculate the arithmetic mean of an ensemble of test results, which is based on the assumption that all experiments were conducted under identical noiseless test conditions. For most room acoustic measurement scenarios, this assumption is not valid, and non-stationary sources of noise often contaminate the results. Traditional statistical averaging methods can be improved by explicitly modeling the ambient interference and noise. Using a signal model for the noise and interference, the proposed parameter estimation procedure provides more accurate results than simple averaging in low signal to noise ratio test scenarios. This method, in which multiple, low volume measurements replace high volume test signals, provides a practical and cost-effective approach for characterizing acoustical spaces.