Subjective quality of the same speech transmission conditions in seven different countries

Abstract

Because of the diversity of subjective testing methods, it is not possible, in general, to compare speech quality measurements from different laboratories. A recent experiment had the aim of determining whether comparable results can be obtained when the same test is performed in several different countries. Tape recordings of speech samples in the native languages of Britain, Canada, France, Italy, Japan, Norway, and the United States were processed by 38 communications circuits in the Communications Methods Research Department of Bell Laboratories. Recordings of the processed speech were returned to each location for evaluation, on a five-point scale, by native listeners. The communications circuits included white noise and speechmodulated noise references, pulse code modulation (PCM) references and four adaptive differential PCM (ADPCM) test circuits, with bit rates between 16 and 48 kbits/s, each with four binary error rates. An important part of the data analysis was estimation of the quality of the 16 ADPCM conditions at location B , given measurements of ADPCM quality at location A and knowledge of reference-circuit quality measured at A and at B . The subjective testing data indicate that a reasonably accurate estimate of ADPCM quality at location B is the quality measured at A adjusted by an additive constant which is the difference between average reference-circuit measurements at B and at A . The accuracy of this estimate is equivalent to that of a subjective test at location B with about seven evaluations of each transmission condition. Three physical measurements are less reliable guides to subjective quality. They lead to estimates with less accuracy than the average of three subjective ratings per circuit.