Military Speech Communications over Vocoders in Tandem

摘要

Speech intelligibility of two types of vocoders was measured using the modified rhyme test. One type of vocoder, a continuous variable slope delta (CVSD), was a waveform encoder. The other type, an advanced multi-band excitation (AMBE), was a parametric encoder. In the first experiment, clear speech was processed through the vocoders. Intelligibility was measured in a control condition, i.e. without vocoding, with each type alone and with two vocoders in tandem. AMBE and CVSD performed similarly, 92.6 and 90.4%, respectively. CVSD-to-AMBE had little effect on intelligibility, measured at 89.2%. However, AMBEto- CVSD had a large degrading effect on intelligibility. The AMBE-to-CVSD direction scored about 81.7% intelligibility with clear, unaltered speech signals. The asymmetry between waveform-to-parametric and parametric-to-waveform encoders underscores the non-linear nature of tandem vocoders on intelligibility. When vocoders of the same type were in tandem, there was no additional effect on intelligibility. The double CVSD condition yielded 92.2% intelligibility and the double AMBE condition yielded 91%. The deleterious effects of speech clipping were measured in a second experiment, as these are ubiquitous in military radio transmission systems. The AMBE parametric vocoder performed at the 88% level in isolation and at 84% when tandemed with the CVSD waveform vocoder. Alternative methods of encoding speech signals are being explored to improve speech intelligibility performance in military communication systems.