# HG changeset patch
# User Chris Cannam
# Date 1400232526 -3600
# Node ID e1ebbae52cff8d18e5faed0e848922de0ca29ef4
# Parent  c6b6d6ea1c3d312c53ccf250b283e79d7a6b29e6
Note on parameters etc

diff -r c6b6d6ea1c3d -r e1ebbae52cff .hgsubstate
--- a/.hgsubstate	Fri May 16 10:12:55 2014 +0100
+++ b/.hgsubstate	Fri May 16 10:28:46 2014 +0100
@@ -1,2 +1,2 @@
 a2621605a8c2119bb063c59e83f4fe792a58e4ac bqvec
-81484e7a33bc1215b09941bc891a560f07973ddc constant-q-cpp
+f3deb9eb13f531ae54f2cb35f856578dc7761b5b constant-q-cpp
diff -r c6b6d6ea1c3d -r e1ebbae52cff src/Silvet.cpp
--- a/src/Silvet.cpp	Fri May 16 10:12:55 2014 +0100
+++ b/src/Silvet.cpp	Fri May 16 10:28:46 2014 +0100
@@ -330,11 +330,14 @@
                         processingSampleRate / 3,
                         processingBPO);
 
-    params.q = 1.0; // MIREX code uses 0.8, but for some reason that
-                    // makes our implementation much, much slower
+    params.q = 0.95; // MIREX code uses 0.8, but it seems 0.9 or lower
+                     // drops the FFT size to 512 from 1024 and alters
+                     // some other processing parameters, making
+                     // everything much, much slower. Could be a flaw
+                     // in the CQ parameter calculations, must check
     params.atomHopFactor = 0.3;
     params.threshold = 0.0005;
-    params.window = CQParameters::Hann;
+    params.window = CQParameters::Hann; //!!! todo: test whether it makes any difference
 
     m_cq = new CQSpectrogram(params, CQSpectrogram::InterpolateLinear);