Skip to main content

Table 5 WER [%] for LVCSR task with 340k bigram LM

From: Advanced acoustic modelling techniques in MP3 speech recognition

Features Raw 128k 32k 28k 24k 20k 16k 12k
PLP_base 23.74 24.5 24.5 25.57 25.98 28.19 38.79 68.57
PLP_adapt 18.19 19.45 19.4 19.59 19.84 20.67 23.56 33.19
PLP_MMI 14.25 14.43 14.55 14.54 15.21 16.15 18.57 25.23
MFCC_base 23.72 25.07 25.13 26.67 31.75 38.46 62.45 91.43
MFCC_adapt 18.44 19.06 19.11 19.92 20.7 22.51 28.11 44.82
MFCC_MMI 14.22 14.72 14.92 15.12 15.82 17.57 21.48 31.54
dMFCC_base 24.85 25.05 26.01 31.77 36.69 48.83 70.03
dMFCC_adapt 18.75 19.01 19.61 20.32 21.71 25.17 34.75
dMFCC_MMI 14.25 14.78 15.06 15.5 16.84 19.47 26.41