Table 2 PER [%] for MFCC-based system for progressively refined AM
From: Advanced acoustic modelling techniques in MP3 speech recognition
Data | Baseline | LDA | SAT | SGMM | MMI |
|---|---|---|---|---|---|
Raw | 17.8 | 15.9 | 12.3 | 10.3 | 7.3 |
128k | 18.9 | 17.2 | 13.7 | 10.8 | 8.1 |
32k | 20.9 | 17.9 | 13.6 | 11.3 | 8.7 |
28k | 24.6 | 19.5 | 14.3 | 11.9 | 9.4 |
24k | 33.9 | 25.9 | 16.4 | 13.5 | 11.4 |
20k | 47.1 | 31.9 | 19.0 | 15.6 | 13.6 |
16k | 62.2 | 49.4 | 25.8 | 20.4 | 18.7 |
12k | 73.0 | 68.7 | 46.0 | 32.6 | 30.2 |