public DefaultFingerprintingConfiguration()
{
// The parameters used in these transformation steps will be equal to those that have been found to work well in other audio fingerprinting studies
// (specifically in A Highly Robust Audio Fingerprinting System):
// audio frames that are 371 ms long (2048 samples),
// taken every 11.6 ms (64 samples),
// thus having an overlap of 31/32
FingerprintLength = 128;
WdftSize = 2048; // 2048/5512 = 371 ms
Overlap = 64; // 64/5512 = 11,6 ms
SamplesPerFingerprint = FingerprintLength * Overlap;
MinFrequency = 318;
MaxFrequency = 2000;
SampleRate = 5512;
LogBase = 2; // 2 or 10;
// In Content Fingerprinting Using Wavelets, a static 928 ms stride was used in database creation,
// and a random 0-46 ms stride was used in querying (random stride was used in order to minimize the coarse effect of unlucky time alignment).
Stride = new IncrementalStaticStride(5115, FingerprintLength * Overlap); // 5115 / 5512 = 0,928 sec
TopWavelets = 200;
LogBins = 32;
WindowFunction = new HannWindow(WdftSize);
NormalizeSignal = true;
UseDynamicLogBase = false;
}
public FullFrequencyFingerprintingConfiguration(bool useRandomStride = false)
{
// http://www.codeproject.com/Articles/206507/Duplicates-detector-via-audio-fingerprinting
// The parameters used in these transformation steps will be close to those that have been found to work well in other audio fingerprinting studies
// (specifically in A Highly Robust Audio Fingerprinting System):
// audio frames that are 371 ms long
// taken every 11.6 ms,
// thus having an overlap of 31/32
// 371 ms is 2048/5512 or 16384/44100 or 11889/32000
// The closest power of 2 in 2's complement format: 8192 / 32000 = 256 ms
WindowSize = 8192;
// 11,6 ms is 64/5512 or 512/44100 or 372/32000
// The closest power of 2 in 2's complement format: 512 / 32000 = 16 ms
// 1024 / 32000 = 32 ms
Overlap = 1024;
// Gets number of samples to read in order to create single signature.
// The granularity is 1.48 seconds (11,6 ms * 128) for SR 5512 hz
// The granularity is 2.048 seconds (16 ms * 128) for SR 32000 hz
// 512 * 128 = 65536
FingerprintLength = 128;
SamplesPerFingerprint = FingerprintLength * Overlap;
// (Originally this was 32, but 40 seems to work better with SCMS?!)
LogBins = 40;
// Each fingerprint will be LogBins x FingerprintLength x 2 Bits long
// e.g. 128 x 32 x 2 = 8192
// or 128 x 40 x 2 = 10240
StartFingerprintIndex = 0;
EndFingerprintIndex = LogBins * FingerprintLength * 2;
// Reduce the frequency range
MinFrequency = 40; // 318; Full Frequency: 20
MaxFrequency = 16000; // 2000; Full Frequency: 22050
// Using 32000 (instead of 44100) gives us a max of 16 khz resolution, which is OK for normal adult human hearing
SampleRate = 32000; // 5512 or 44100
LogBase = 10; // Math.E, 2 or 10;
// In Content Fingerprinting Using Wavelets, a static 928 ms stride was used in database creation,
// and a random 0-46 ms stride was used in querying (random stride was used in order to minimize the coarse effect of unlucky time alignment).
if (useRandomStride) {
// 0,046 sec is 2028 / 44100 or 1472/32000
// use a 128 ms random stride instead = 4096, since every 46 ms gives way too many fingerprints to query efficiently
Stride = new IncrementalRandomStride(1, 4096, SamplesPerFingerprint);
} else {
// 0,928 sec is 5115 / 5512 or 40924 / 44100 or 29695/32000
Stride = new IncrementalStaticStride(29695, SamplesPerFingerprint);
}
TopWavelets = 200;
WindowFunction = new HannWindow(WindowSize);
NormalizeSignal = true; // true;
UseDynamicLogBase = false; // false;
// Number of LSH tables
NumberOfHashTables = 25;
// Number of Min Hash keys per 1 hash function (1 LSH table)
NumberOfKeys = 4;
}
