A noise annoys an oyster: evaluating the subjective effectiveness of single-input noise reduction techniques

Single-input noise reduction techniques show enormous potential for improving the quality of signals that originate in noisy environments or that become mixed with noise along the way. These techniques are of particular interest to developers of speech compression codecs since the quality of noise processed through the codec is often a limiting factor for codec performance. This paper considers the assessment of the subjective performance of noise reduction (NR) techniques for telephony applications. Parameters of interest include the selection of the source material (such as noise type and level), voice characteristics, input impairments, the presence of other DSP processing on the channel, and the choice of test procedure. Some results from absolute category rating tests are presented, showing the subjective improvement found with NR techniques based on spectral subtraction.