v3.1, 20040330, ME
Session P Tuesday, May 11 09:30 h–12:30 h
PSYCHOACOUSTICS, PERCEPTION, AND LISTENING TESTS
Chair: Søren Bech, Bang & Olufsen a/s, Denmark
P-1 Quality Adviser: A Multichannel Audio Quality Expert System—Slawomir Zielinski1, Francis Rumsey1, Rafael Kassier1, Søren Bech2
1 University of Surrey, Guildford, Surrey, UK
2 Bang & Olufsen a/s, Streuer, Denmark
The basic audio quality of 5.1 multichannel audio reproduction was evaluated under different technical conditions. The obtained database of subjective responses was used to develop a multichannel audio quality expert system. There are three aims of this development: (1) to predict audio quality as a function of individual channel bandwidth, (2) to predict audio quality as a function of a down-mix algorithm, (3) to predict the optimum technical trade-off between these factors for a given overall bandwidth of a multichannel audio signal. Obtained results indicate a close correspondence between the predicted and actual quality ratings. It is intended that the final version of the Quality Adviser will be suitable as a decision making aid for broadcasters and codec designers.
P-2 Subjective Evaluation of Virtual Home Theater Sound Systems for Loudspeakers and Headphones—Gaëtan Lorho, Nick Zacharov, Nokia Research Center, Tampere, Finland
A subjective evaluation of Virtual Home Theater systems (VHT) for loudspeaker and headphone reproduction is presented in this paper. Several algorithms for loudspeakers and headphones were selected and applied to six different multichannel audio programs. A subjective experiment was performed for each configuration using screened listeners to assess the performance of these VHT algorithms in terms of overall sound reproduction quality. A paired comparison method was chosen, with the discrete 5-channel reproduction (3/2) system as a reference in the loudspeaker test, and the stereo downmix of the 5-channel material in the headphone test. The stereo downmix was also compared to the 5-channel reference in the case of the loudspeaker reproduction. The experimental design and the detailed analysis of results are presented in this paper.
P-3 Elicitation and Grading of Subjective Attributes of 2-Channel Phantom Images—Hyun-Kook Lee, Francis Rumsey, University of Surrey, Guildford, Surrey, UK
The subjective attributes of 2-channel phantom images of transient piano, continuous trumpet, and male speech sources were elicited using pair-wise comparison between reference mono images and their phantom images. The attributes elicited included image focus, image width, image distance, brightness, hardness, and fullness. The effect of interchannel time and intensity differences on the perceived difference between the real image and its phantom image was investigated for each sound source with respect to the elicited subjective attributes. Results show that the type of panning method (pure time, pure intensity, and combination of the two) had a statistically significant effect on image focus and image width attributes. It was also found that the type of sound source had a significant effect on all the attributes.
P-4 Loudness Assessment of Music and Speech—Esben Skovenborg1, 3, René Quesnel2, Søren H. Nielsen1
1 TC Electronic A/S, Risskov, Denmark
2 McGill University, Montreal, Quebec, Canada
3 University of Aarhus, Aarhus, Denmark
An experiment was performed to investigate the assessment of loudness of music and speech using a general linear model. Eight expert listeners participated in the experiment. The method of adjustment was used for loudness matching of stimuli. Both stimuli of each pair was selected from a collection of 147 homogeneous audio segments including representative samples of speech, jazz, rock/pop, and classical music, together with pink noise and a 1-kHz tone. For each segment, a reliable estimate of the loudness level was obtained from the model. Both the uncertainty and the subjectivity factors were shown to depend on the category of the stimuli. An alternative categorization based on four MPEG-7 audio descriptors was also used for the analysis.
P-5 Imperfections at Low Frequencies—How Audible or Annoying Are They?—Tomas Salava, ETOS acoustics, Prague, Czech Republic
This paper deals with some open problems of low-frequency sound reproduction, particularly in medium and small listening rooms. First, the basic facts concerning sound fields and transfer functions in bounded spaces are briefly recalled. Specifics of sound quality perception at low frequencies are then outlined. Opinion differences in this field are discussed too. Strong influence of the test signals properties is stressed, and using both musical, and artificial test signals for low-frequency listening tests is recommended. Several examples of different artificial low-frequency test signals are described and compared with musical signals.
P-6 New Intrusive Method for the Objective Quality Evaluation of Acoustic Noise Suppression in Mobile Communications—Juha Salmela, Ville-Veikko Mattila, Nokia Research Center, Tampere Finland
A new intrusive method, combining several independent objective metrics, has been developed for the evaluation of the quality of acoustic noise suppression in mobile communications. Extensive subjective data, including simulations of several noise suppression solutions in various noise environments, was gathered to serve as the benchmark for the metrics. Partial least-square regression and full cross-validation were used to establish the applicability of 26 metrics, which were making use of different measurement procedures, to predict the perceived quality. A Phase IV, vector-based preference model was optimized to predict quality with a correlation of 0.95, resulting in an average prediction error of 8 percent. Different measurement procedures appeared to contribute to a similar extent to the prediction ability of the optimized model.
