|
|
 |
An Independent Replication of The Diagnostic Acceptability Measure (DAM) ABSTRACT For the first time, an independent replication of the DAM [W.D. Voiers, IEEE Proc. ICASSP 204-207 (1977)], a Dynastat proprietary test, was conducted as part of a three phase effort to facilitate secondary sources for the DAM. Dynastat conducted phases one and two of this effort. In the first phase, a description of the DAM and the procedures to conduct the test were developed. In the second phase, an extensive set of speech distortions was evaluated with the DAM at Dynastat. For this study, the results of the initial phase were used to set up a test facility and to recruit, screen, train and calibrate a crew of listeners. This independent facility and listener crew then used the DAM for evaluation of the same set of speech distortions tested at Dynastat during phase two of the overall effort. The distortion set included various types of masking noise, filtering, amplitude distortions, interrupts and signal compression algorithms. Most distortion types were tested at six levels designed to cover the measurement range of the DAM. There were a total of 180 conditions tested for six speakers each. Initial comparative analysis between the DAM results (Composite Acceptability Measure, CAE, averaged over speakers and listeners) for Dynastat and this study demonstrated a strong linear relationship (r=.97) and a standard error of the estimate equal to 3.3 DAM points. Outliers to this regression analysis were grouped by some distortion types. In order to study intra-listener variability of the two crews, the isometric and parametric components of the composite measure, and distortion type variability, further analyses were conducted. A modified significant difference method was used on the overall and component measures for the individual conditions. These analyses led to recommendations for additions to the phase one DAM description and to the DAM listener calibration process. [Work supported by USAF Rome Laboratory RL/EEV.] |