Skip to main content

Table 4 Results for dithered MFCCs, the dithering value, and the corresponding error rate R/PER [%]

From: Advanced acoustic modelling techniques in MP3 speech recognition

Data Baseline LDA SAT SGMM MMI
128k 2/18.9 2/17.1 4/13.1 4/10.7 4/8.0
32k 8/20.1 4/17.8 4/13.4 4/11.1 2/8.6
28k 8/21.6 4/18.6 8/13.9 8/11.5 8/9.1
24k 16/30.1 2/26.2 8/15.8 8/13.1 8/10.9
20k 16/34.7 8/30.0 8/17.7 8/14.8 8/12.8
16k 32/41.2 32/35.9 16/21.3 8/17.9 8/16.4
12k 64/49.9 64/45.5 16/29.0 16/24.2 16/23.0