Results of experiment 1 for the six-channel vocoder (dashed line) and the 11-channel vocoder (solid line). Error bars show SD. The target and background were subject to multi-channel compression either AFTER or BEFORE mixing. Comparisons of performance across channel numbers are not meaningful, as the TBR differed for the six- and 11-channel systems.
Plot of fractional modulation reduction, , for the four compressor speeds used in experiment 2 (solid lines labeled 1–4; compression speed increases progressively from 1 to 4). For comparison purposes, the dashed lines marked S and F are for the slow and fast compressors, respectively, used by Stone and Moore (2003). The numbers 1.0, 2.0 and 4.0 plotted adjacent to the grid line indicate effective compression ratios.
Results of experiment 2: the dashed lines show results for eight channels and the solid lines show results for 12 channels. Comparisons of performance across channel numbers are not meaningful, as the TBR differed for the eight- and 12-channel systems. The error bars show SD. Note the broken scale and different values on the ordinate between the results for the reference experiment (upper lines) and main experiment (lower lines). For system , no compression was applied; plots for compression systems 1–4 are shown in Fig. 2.
Plots of versus modulation frequency for the four compressor speeds used in experiment 3 (solid lines, labeled 1–4). For comparison purposes, the dashed lines marked S and F are for the slow and fast compressors, respectively, used by Stone and Moore (2003).
Results of experiment 3 represented as contours of equal intelligibility. The left-hand panel shows results for the 12-channel vocoder and the right-hand panel shows results for the 18-channel vocoder. The abscissa shows the compressor speed, as defined in Fig. 4, while the ordinate shows the number of channels of independent compression (logarithmically spaced).
Contour plots of the measures WSMC, FES and ASMC, plotted in a similar format to that for Fig. 5. The upper row shows plots for the 12-channel vocoder and the lower row shows plots for the 18-channel vocoder. The “diff” values indicate the difference between the highest and lowest values in each contour plot. The contours are spaced at 0.2 times the “diff” values.
Compression ratios (CRs) used in the channel compressors of experiment 1. Part A is for the six-channel system, which used a TBR of . Part B is for the 11-channel system, which used a TBR of . The first line in each section shows the channel index. Lines two and three show the static CR and the corresponding value of , the fractional reduction in modulation, that was set for each channel of the system with compression BEFORE mixing, in order to match the effective compression in each channel to that produced by the single-channel compressor, whose CR was 2.78. Lines four and five show CRs for the system with compression AFTER mixing and the corresponding values.
Values of the various measures of envelope distortion for the stimuli used in experiment 1. The bottom row shows the mean intelligibility scores for these stimuli. Intelligibility was not measured for the unprocessed (uncompressed) stimuli.
Comparison between the scores (in percent correct) of experiment 2 of Stone and Moore (2004) and experiment 1 here. Part A is for the six-channel system. Part B is for the 11-channel system. The second line in each section shows the scores in experiment 2 of Stone and Moore (2004) which used single-channel compression. Line three shows the scores in experiment 1 here which used multi-channel compression. The final column shows the difference in scores between conditions INDEP/BEFORE and COMOD/AFTER.
Edge frequencies in Hz for the band-pass filters used to create channels in experiments 2 and 3. For experiment 2 (rows marked SII), edge frequencies were chosen to produce an equal contribution to the SII (ANSI, 1997) from each band, and compression channel edge frequencies equaled vocoder channel edge frequencies. For experiment 3, the compression processing used 1, 3, 6, or 12 channels while the noise vocoder used 12 or 18 channels, and edge frequencies had equal spacing on the -number scale (Glasberg and Moore, 1990).
Values of the various measures of envelope distortion for the stimuli used in experiment 2. The bottom row of each section of the table shows the mean intelligibility scores. N indicates the condition with no compression. The numbers 1–4 are labels for the different compression speeds.
Scores in percent correct for experiment 3 for each combination of compressor speed and number of compression channels. Part A is for the 12-channel vocoder and part B is for the 18-channel vocoder. Numbers in parentheses are SDs. Standard errors would be a factor of 5.6 smaller.
Article metrics loading...
Full text loading...