summary | shortlog | log | commit | commitdiff | tree
raw | patch | inline | side by side (parent: 6d885ee)
raw | patch | inline | side by side (parent: 6d885ee)
author | Wonkyum <wonkyuml@users.noreply.github.com> | |
Thu, 24 Aug 2017 05:57:30 +0000 (22:57 -0700) | ||
committer | Daniel Povey <dpovey@gmail.com> | |
Thu, 24 Aug 2017 05:57:30 +0000 (22:57 -0700) |
egs/swbd/s5c/conf/mfcc_hires_bp.conf | [new file with mode: 0644] | patch | blob |
diff --git a/egs/swbd/s5c/conf/mfcc_hires_bp.conf b/egs/swbd/s5c/conf/mfcc_hires_bp.conf
--- /dev/null
@@ -0,0 +1,13 @@
+# config for high-resolution MFCC features, intended for speech activity detection.
+# This differs from conf/mfcc_hires.conf, since it is bandlimited between
+# 330 Hz and 3000 Hz, the frequencies corresponding to the telephone
+# bandwidth.
+# Note: we keep all cepstra, so it has the same info as filterbank features,
+# but MFCC is more easily compressible (because less correlated) which is why
+# we prefer this method.
+--use-energy=false # use average of log energy, not energy.
+--sample-frequency=8000 # Babel is sampled at 8kHz
+--num-mel-bins=28
+--num-ceps=28 # there is no dimensionality reduction.
+--low-freq=330 # low cutoff frequency for mel bins
+--high-freq=3000 # high cutoff frequency