public List<bool[]> CreateFingerprintsFromAudioSamples(float[] samples, WorkUnitParameterObject param, out double[][] logSpectrogram, out List<double[][]> spectralImages)
{
IFingerprintingConfiguration configuration = param.FingerprintingConfiguration;
AudioServiceConfiguration audioServiceConfiguration = new AudioServiceConfiguration
{
LogBins = configuration.LogBins,
LogBase = configuration.LogBase,
MaxFrequency = configuration.MaxFrequency,
MinFrequency = configuration.MinFrequency,
Overlap = configuration.Overlap,
SampleRate = configuration.SampleRate,
WindowSize = configuration.WindowSize,
NormalizeSignal = configuration.NormalizeSignal,
UseDynamicLogBase = configuration.UseDynamicLogBase
};
// store the log spectrogram in the out variable
logSpectrogram = AudioService.CreateLogSpectrogram(
samples, configuration.WindowFunction, audioServiceConfiguration);
return this.CreateFingerprintsFromLogSpectrum(
logSpectrogram,
configuration.Stride,
configuration.FingerprintLength,
configuration.Overlap,
configuration.TopWavelets,
out spectralImages);
}
private List<bool[]> CreateFingerprintsFromAudioSamples(float[] samples, WorkUnitParameterObject param)
{
IFingerprintingConfiguration configuration = param.FingerprintingConfiguration;
AudioServiceConfiguration audioServiceConfiguration = new AudioServiceConfiguration
{
LogBins = configuration.LogBins,
LogBase = configuration.LogBase,
MaxFrequency = configuration.MaxFrequency,
MinFrequency = configuration.MinFrequency,
Overlap = configuration.Overlap,
SampleRate = configuration.SampleRate,
WdftSize = configuration.WdftSize
};
float[][] spectrum = audioService.CreateLogSpectrogram(
samples, configuration.WindowFunction, audioServiceConfiguration);
return CreateFingerprintsFromSpectrum(
spectrum,
configuration.Stride,
configuration.FingerprintLength,
configuration.Overlap,
configuration.LogBins,
configuration.TopWavelets);
}
public Task<List<bool[]>> Process(WorkUnitParameterObject details)
{
if (!string.IsNullOrEmpty(details.PathToAudioFile))
{
return Task.Factory.StartNew(() => CreateFingerprintsFromAudioFile(details));
}
return Task.Factory.StartNew(() => CreateFingerprintsFromAudioSamples(details.AudioSamples, details));
}
public List<bool[]> CreateFingerprintsFromAudioFile(WorkUnitParameterObject param, out double[][] logSpectrogram)
{
float[] samples = AudioService.ReadMonoFromFile(
param.PathToAudioFile,
param.FingerprintingConfiguration.SampleRate,
param.MillisecondsToProcess,
param.StartAtMilliseconds);
return CreateFingerprintsFromAudioSamples(samples, param, out logSpectrogram);
}
public Dictionary<Track, QueryStats> FindSimilarFromAudioFile(
int lshHashTables,
int lshGroupsPerKey,
int thresholdTables,
WorkUnitParameterObject param)
{
// Get fingerprints
// TODO: Note that this method might return too few samples
double[][] LogSpectrogram;
List<bool[]> signatures = fingerprintService.CreateFingerprintsFromAudioFile(param, out LogSpectrogram);
long elapsedMiliseconds = 0;
// Query the database using Min Hash
Dictionary<int, QueryStats> allCandidates = QueryFingerprintManager.QueryOneSongMinHash(
signatures,
dbService,
minHash,
lshHashTables,
lshGroupsPerKey,
thresholdTables,
ref elapsedMiliseconds);
IEnumerable<int> ids = allCandidates.Select(p => p.Key);
IList<Track> tracks = dbService.ReadTrackById(ids);
// Order by Hamming Similarity
// Using PLINQ
OrderedParallelQuery<KeyValuePair<int, QueryStats>> order = allCandidates.AsParallel()
//IOrderedEnumerable<KeyValuePair<int, QueryStats>> order = allCandidates
.OrderBy((pair) => pair.Value.OrderingValue =
pair.Value.HammingDistance / pair.Value.NumberOfTotalTableVotes
+ 0.4 * pair.Value.MinHammingDistance);
// Join on the ID properties.
var joined = from o in order
join track in tracks on o.Key equals track.Id
select new { track, o.Value };
Dictionary<Track, QueryStats> stats = joined.ToDictionary(Key => Key.track, Value => Value.Value);
return stats;
}
private static bool AnalyseAndAddScmsUsingFingerprints(List<double[][]> spectralImages,
List<bool[]> fingerprints,
WorkUnitParameterObject param,
Db db,
int trackId,
bool doOutputDebugInfo=DEFAULT_DEBUG_INFO)
{
DbgTimer t = new DbgTimer();
t.Start ();
// Insert Statistical Cluster Model Similarity Audio Feature
string fileName = param.FileName;
// Merge the arrays in the List using Linq
var result = spectralImages.SelectMany(i => i).ToArray();
Comirva.Audio.Util.Maths.Matrix scmsMatrix = new Comirva.Audio.Util.Maths.Matrix(result);
if (doOutputDebugInfo) {
scmsMatrix.DrawMatrixImage(String.Format("{0}_spectral.png", fileName));
}
#region Store in a Statistical Cluster Model Similarity class.
Scms audioFeature = Scms.GetScms(scmsMatrix, fileName);
if (audioFeature != null) {
// Store bitstring hash as well
audioFeature.BitString = GetBitString(scmsMatrix);
// Store duration
audioFeature.Duration = (long) param.DurationInMs;
// Store file name
audioFeature.Name = param.PathToAudioFile;
// Add to database
int id = trackId;
if (db.AddTrack(audioFeature) == -1) {
Console.Out.WriteLine("Failed! Could not add audio feature to database ({0})!", fileName);
return false;
}
} else {
return false;
}
#endregion
Dbg.WriteLine ("AnalyseAndAddScmsUsingFingerprints2 - Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return true;
}
/// <summary>
/// Insert track into database
/// </summary>
/// <param name="track">Track</param>
/// <param name="hashTables">Number of hash tables (e.g. 25)</param>
/// <param name="hashKeys">Number of hash keys (e.g. 4)</param>
/// <param name="param">WorkUnitParameterObject parameters</param>
public bool InsertTrackInDatabaseUsingSamples(Track track, int hashTables, int hashKeys, WorkUnitParameterObject param, out double[][] logSpectrogram)
{
if (dbService.InsertTrack(track)) {
List<bool[]> images = fingerprintService.CreateFingerprintsFromAudioSamples(param.AudioSamples, param, out logSpectrogram);
List<Fingerprint> inserted = AssociateFingerprintsToTrack(images, track.Id);
if (dbService.InsertFingerprint(inserted)) {
return HashFingerprintsUsingMinHash(inserted, track, hashTables, hashKeys);
} else {
return false;
}
} else {
logSpectrogram = null;
return false;
}
}
// TODO: Rememeber to use another stride when querying
public static Dictionary<Track, double> SimilarTracksSoundfingerprinting(FileInfo filePath)
{
DbgTimer t = new DbgTimer();
t.Start ();
FindSimilar.AudioProxies.BassProxy bass = FindSimilar.AudioProxies.BassProxy.Instance;
float[] audiodata = AudioFileReader.Decode(filePath.FullName, SAMPLING_RATE, SECONDS_TO_ANALYZE);
if (audiodata == null || audiodata.Length == 0) {
Dbg.WriteLine("Error! - No Audio Found");
return null;
}
// Name of file being processed
string name = StringUtils.RemoveNonAsciiCharacters(Path.GetFileNameWithoutExtension(filePath.Name));
// Calculate duration in ms
double duration = (double) audiodata.Length / SAMPLING_RATE * 1000;
// Explode samples to the range of 16 bit shorts (–32,768 to 32,767)
// Matlab multiplies with 2^15 (32768)
// e.g. if( max(abs(speech))<=1 ), speech = speech * 2^15; end;
MathUtils.Multiply(ref audiodata, AUDIO_MULTIPLIER); // 65536
// zero pad if the audio file is too short to perform a mfcc
if (audiodata.Length < (fingerprintingConfig.WdftSize + fingerprintingConfig.Overlap))
{
int lenNew = fingerprintingConfig.WdftSize + fingerprintingConfig.Overlap;
Array.Resize<float>(ref audiodata, lenNew);
}
// Get fingerprint signatures using the Soundfingerprinting methods
// Get database
DatabaseService databaseService = DatabaseService.Instance;
IPermutations permutations = new LocalPermutations("Soundfingerprinting\\perms.csv", ",");
Repository repository = new Repository(permutations, databaseService, fingerprintService);
// work config
WorkUnitParameterObject param = new WorkUnitParameterObject();
param.FingerprintingConfiguration = fingerprintingConfig;
param.PathToAudioFile = filePath.FullName;
param.AudioSamples = audiodata;
param.MillisecondsToProcess = SECONDS_TO_ANALYZE * 1000;
param.StartAtMilliseconds = 0;
Dictionary<Track, double> candidates = repository.FindSimilarFromAudioSamples(25, 4, 2, param);
return candidates;
/*
// Use var keyword to enumerate dictionary
foreach (var pair in candidates)
{
Console.WriteLine("{0} - {1:0.00}",
pair.Key.Title,
pair.Value);
}
*/
Dbg.WriteLine ("Soundfingerprinting - Total Execution Time: {0} ms", t.Stop().TotalMilliseconds);
}
/// <summary>
/// Find Similar Tracks using passed audio samples as input and return a List
/// </summary>
/// <param name="lshHashTables">Number of hash tables from the database</param>
/// <param name="lshGroupsPerKey">Number of groups per hash table</param>
/// <param name="thresholdTables">Minimum number of hash tables that must be found for one signature to be considered a candidate (0 and 1 = return all candidates, 2+ = return only exact matches)</param>
/// <param name="param">Audio File Work Unit Parameter Object</param>
/// <param name="optimizeSignatureCount">Reduce the number of signatures in order to increase the search performance</param>
/// <param name="doSearchEverything">disregard the local sensitivity hashes and search the whole database</param>
/// <param name="splashScreen">The "please wait" splash screen (or null)</param>
/// <returns>a list of perceptually similar tracks</returns>
public List<FindSimilar.QueryResult> FindSimilarFromAudioSamplesList(
int lshHashTables,
int lshGroupsPerKey,
int thresholdTables,
WorkUnitParameterObject param,
bool optimizeSignatureCount,
bool doSearchEverything,
SplashSceenWaitingForm splashScreen) {
DbgTimer t = new DbgTimer();
t.Start ();
if (splashScreen != null) splashScreen.SetProgress(2, "Creating fingerprints from audio samples ...");
// Get fingerprints
double[][] logSpectrogram;
List<bool[]> fingerprints = fingerprintService.CreateFingerprintsFromAudioSamples(param.AudioSamples, param, out logSpectrogram);
#if DEBUG
// Save debug images using fingerprinting methods
//Analyzer.SaveFingerprintingDebugImages(param.FileName, logSpectrogram, fingerprints, fingerprintService, param.FingerprintingConfiguration);
#endif
if (splashScreen != null) splashScreen.SetProgress(3, String.Format("Successfully created {0} fingerprints.", fingerprints.Count));
// If the number of signatures is to big, only keep the first MAX_SIGNATURE_COUNT to avoid a very time consuming search
if (optimizeSignatureCount && fingerprints.Count > MAX_SIGNATURE_COUNT) {
if (splashScreen != null) splashScreen.SetProgress(4, String.Format("Only using the first {0} fingerprints out of {1}.", MAX_SIGNATURE_COUNT, fingerprints.Count));
fingerprints.RemoveRange(MAX_SIGNATURE_COUNT, fingerprints.Count - MAX_SIGNATURE_COUNT);
Dbg.WriteLine("Only using the first {0} fingerprints.", MAX_SIGNATURE_COUNT);
}
long elapsedMiliseconds = 0;
// Query the database using Min Hash
Dictionary<int, QueryStats> allCandidates = QueryFingerprintManager.QueryOneSongMinHash(
fingerprints,
dbService,
minHash,
lshHashTables,
lshGroupsPerKey,
thresholdTables,
ref elapsedMiliseconds,
doSearchEverything,
splashScreen);
if (splashScreen != null) splashScreen.SetProgress(91, String.Format("Found {0} candidates.", allCandidates.Count));
IEnumerable<int> ids = allCandidates.Select(p => p.Key);
IList<Track> tracks = dbService.ReadTrackById(ids);
if (splashScreen != null) splashScreen.SetProgress(95, String.Format("Reading {0} tracks.", tracks.Count));
// Order by Hamming Similarity
// TODO: What does the 0.4 number do here?
// there doesn't seem to be any change using another number?!
/*
// Using PLINQ
IOrderedEnumerable<KeyValuePair<int, QueryStats>> order = allCandidates
.OrderBy((pair) => pair.Value.OrderingValue =
pair.Value.HammingDistance / pair.Value.NumberOfTotalTableVotes
+ 0.4 * pair.Value.MinHammingDistance);
var fingerprintList = (from o in order
join track in tracks on o.Key equals track.Id
select new FindSimilar.QueryResult {
Id = track.Id,
Path = track.FilePath,
Duration = track.TrackLengthMs,
Similarity = o.Value.Similarity
}).ToList();
*/
// http://msdn.microsoft.com/en-us/library/dd460719(v=vs.110).aspx
// http://stackoverflow.com/questions/2767709/c-sharp-joins-where-with-linq-and-lambda
// Lambda query to order the candidates
var order = allCandidates.AsParallel()
.OrderBy((pair) => pair.Value.OrderingValue =
pair.Value.HammingDistance / pair.Value.NumberOfTotalTableVotes
+ 0.4 * pair.Value.MinHammingDistance)
.Take(200);
// TODO: Be able to create the above query as a LINQ query
// Join on the ID properties.
var fingerprintList = (from o in order.AsParallel()
join track in tracks.AsParallel() on o.Key equals track.Id
select new FindSimilar.QueryResult {
Id = track.Id,
Path = track.FilePath,
Duration = track.TrackLengthMs,
Similarity = o.Value.Similarity
})
.OrderByDescending((ord) => ord.Similarity)
.ToList();
if (splashScreen != null) splashScreen.SetProgress(100, "Ready!");
Dbg.WriteLine ("FindSimilarFromAudioSamplesList - Total Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return fingerprintList;
}
/// <summary>
/// Return information from the Audio File
/// </summary>
/// <param name="filePath">filepath object</param>
/// <returns>a WorkUnitParameter object</returns>
public static WorkUnitParameterObject GetWorkUnitParameterObjectFromAudioFile(FileInfo filePath, bool doOutputDebugInfo=DEFAULT_DEBUG_INFO)
{
DbgTimer t = new DbgTimer();
t.Start ();
float[] audiodata = AudioFileReader.Decode(filePath.FullName, SAMPLING_RATE, SECONDS_TO_ANALYZE);
if (audiodata == null || audiodata.Length == 0) {
Dbg.WriteLine("Error! - No Audio Found");
return null;
}
// Name of file being processed
string fileName = StringUtils.RemoveNonAsciiCharacters(Path.GetFileNameWithoutExtension(filePath.Name));
#if DEBUG
if (DEBUG_INFO_VERBOSE) {
if (DEBUG_OUTPUT_TEXT) WriteAscii(audiodata, fileName + "_audiodata.ascii");
if (DEBUG_OUTPUT_TEXT) WriteF3Formatted(audiodata, fileName + "_audiodata.txt");
}
#endif
if (doOutputDebugInfo) {
DrawGraph(MathUtils.FloatToDouble(audiodata), fileName + "_audiodata.png");
}
// Calculate duration in ms
double duration = (double) audiodata.Length / SAMPLING_RATE * 1000;
// Explode samples to the range of 16 bit shorts (–32,768 to 32,767)
// Matlab multiplies with 2^15 (32768)
// e.g. if( max(abs(speech))<=1 ), speech = speech * 2^15; end;
MathUtils.Multiply(ref audiodata, AUDIO_MULTIPLIER);
// zero pad if the audio file is too short to perform a mfcc
if (audiodata.Length < (WINDOW_SIZE + OVERLAP))
{
int lenNew = WINDOW_SIZE + OVERLAP;
Array.Resize<float>(ref audiodata, lenNew);
}
// work config
WorkUnitParameterObject param = new WorkUnitParameterObject();
param.AudioSamples = audiodata;
param.PathToAudioFile = filePath.FullName;
param.MillisecondsToProcess = SECONDS_TO_ANALYZE * 1000;
param.StartAtMilliseconds = 0;
param.FileName = fileName;
param.DurationInMs = duration;
param.Tags = GetTagInfoFromFile(filePath.FullName);
Dbg.WriteLine ("Get Audio File Parameters - Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return param;
}
private static List<int[]> TestSoundfingerprintingAlgorithm(float[] samples, string name)
{
// work config
WorkUnitParameterObject param = new WorkUnitParameterObject();
param.FingerprintingConfiguration = fingerprintingConfig;
// Soundfingerprinting Service
FingerprintService fingerprintService = GetSoundfingerprintingService();
// Image Service
ImageService imageService =
new ImageService(fingerprintService.SpectrumService, fingerprintService.WaveletService);
// Get fingerprints
double[][] LogSpectrogram;
List<bool[]> fingerprints = fingerprintService.CreateFingerprintsFromAudioSamples(samples, param, out LogSpectrogram);
int width = param.FingerprintingConfiguration.FingerprintLength;
int height = param.FingerprintingConfiguration.LogBins;
imageService.GetImageForFingerprints(fingerprints, width, height, 2).Save("imageservice_" + name + "_fingerprints.png");
IPermutations permutations = new LocalPermutations("Soundfingerprinting\\perms.csv", ",");
/*
Soundfingerprinting.DuplicatesDetector.DataAccess.Repository repository = new Soundfingerprinting.DuplicatesDetector.DataAccess.Repository(permutations);
// Define track
Soundfingerprinting.DuplicatesDetector.Model.Track track
= new Soundfingerprinting.DuplicatesDetector.Model.Track {
Title = name
};
*/
Soundfingerprinting.Hashing.MinHash hasher = new Soundfingerprinting.Hashing.MinHash(permutations);
List<int[]> signatures = new List<int[]>();
foreach (bool[] fingerprint in fingerprints)
{
int[] signature = hasher.ComputeMinHashSignature(fingerprint); /*Compute min-hash signature out of signature*/
signatures.Add(signature);
}
return signatures;
}
private static void TestSoundfingerprintingAlgorithm(string filename, string name)
{
// work config
WorkUnitParameterObject param = new WorkUnitParameterObject();
param.PathToAudioFile = filename;
param.StartAtMilliseconds = 0;
param.MillisecondsToProcess = 0;
param.FingerprintingConfiguration = fingerprintingConfig;
// Soundfingerprinting Service
FingerprintService fingerprintService = GetSoundfingerprintingService();
// Image Service
ImageService imageService = new ImageService(
fingerprintService.SpectrumService,
fingerprintService.WaveletService);
// Configuration
AudioServiceConfiguration audioServiceConfiguration = new AudioServiceConfiguration
{
LogBins = fingerprintingConfig.LogBins,
LogBase = fingerprintingConfig.LogBase,
MaxFrequency = fingerprintingConfig.MaxFrequency,
MinFrequency = fingerprintingConfig.MinFrequency,
Overlap = fingerprintingConfig.Overlap,
SampleRate = fingerprintingConfig.SampleRate,
WdftSize = fingerprintingConfig.WdftSize,
NormalizeSignal = fingerprintingConfig.NormalizeSignal,
UseDynamicLogBase = fingerprintingConfig.UseDynamicLogBase
};
double[][] spectrogram = fingerprintService.AudioService.CreateSpectrogram(filename, new Mirage.HannWindow(fingerprintingConfig.WdftSize), fingerprintingConfig.SampleRate, fingerprintingConfig.Overlap, fingerprintingConfig.WdftSize);
imageService.GetSpectrogramImage(spectrogram, 600, 400).Save("imageservice_" + name + "_specgram.png");
/*
Comirva.Audio.Util.Maths.Matrix stftdata = new Comirva.Audio.Util.Maths.Matrix(spectrogram).Transpose();
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE) {
if (DEBUG_OUTPUT_TEXT) {
stftdata.WriteAscii(name + "_stftdata2.ascii");
stftdata.WriteCSV(name + "_stftdata2.csv", ";");
}
// same as specgram(audio*32768, 2048, 44100, hanning(2048), 1024);
stftdata.DrawMatrixImageLogValues(name + "_specgram2.png", true);
// spec gram with log values for the y axis (frequency)
stftdata.DrawMatrixImageLogY(name + "_specgramlog2.png", SAMPLING_RATE, 20, SAMPLING_RATE/2, 120, WINDOW_SIZE);
}
#endif
*/
double[][] logSpectrogram = fingerprintService.AudioService.CreateLogSpectrogram(filename, new Mirage.HannWindow(fingerprintingConfig.WdftSize), audioServiceConfiguration);
imageService.GetLogSpectralImages(logSpectrogram, fingerprintingConfig.Stride, fingerprintingConfig.FingerprintLength, fingerprintingConfig.Overlap, 2).Save("imageservice_" + name + "_specgram_logimages.png");
Comirva.Audio.Util.Maths.Matrix stftdataLog = new Comirva.Audio.Util.Maths.Matrix(logSpectrogram).Transpose();
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE) {
if (DEBUG_OUTPUT_TEXT) {
stftdataLog.WriteAscii(name + "_stftdataLog.ascii");
stftdataLog.WriteCSV(name + "_stftdataLog.csv", ";");
}
// same as specgram(audio*32768, 2048, 44100, hanning(2048), 1024);
stftdataLog.DrawMatrixImageLogValues(name + "_stftdataLog.png", true);
}
#endif
// Get fingerprints
double[][] LogSpectrogram;
List<bool[]> fingerprints = fingerprintService.CreateFingerprintsFromAudioFile(param, out LogSpectrogram);
int width = fingerprintingConfig.FingerprintLength;
int height = fingerprintingConfig.LogBins;
imageService.GetImageForFingerprints(fingerprints, width, height, 2).Save("imageservice_" + name + "_fingerprints.png");
/*
IPermutations permutations = new LocalPermutations("Soundfingerprinting\\perms.csv", ",");
Soundfingerprinting.DuplicatesDetector.DataAccess.Repository repository = new Soundfingerprinting.DuplicatesDetector.DataAccess.Repository(permutations);
// Define track
Soundfingerprinting.DuplicatesDetector.Model.Track track
= new Soundfingerprinting.DuplicatesDetector.Model.Track {
Title = name,
Path = filename
};
// Get the HashSignatures
List<Soundfingerprinting.DuplicatesDetector.Model.HashSignature> signatures = repository.GetSignatures(fingerprints, track, 25, 4);
return signatures;
*/
}
/// <summary>
/// Find Similar Tracks using passed audio samples as input
/// </summary>
/// <param name="lshHashTables">Number of hash tables from the database</param>
/// <param name="lshGroupsPerKey">Number of groups per hash table</param>
/// <param name="thresholdTables">Threshold percentage [0.07 for 20 LHash Tables, 0.17 for 25 LHashTables]</param>
/// <param name="param">Audio File Work Unit Parameter Object</param>
/// <returns>a list of perceptually similar tracks</returns>
public List<FindSimilar.QueryResult> FindSimilarFromAudioSamplesList(
int lshHashTables,
int lshGroupsPerKey,
int thresholdTables,
WorkUnitParameterObject param)
{
// Get fingerprints
double[][] logSpectrogram;
List<double[][]> spectralImages;
List<bool[]> signatures = fingerprintService.CreateFingerprintsFromAudioSamples(param.AudioSamples, param, out logSpectrogram, out spectralImages);
long elapsedMiliseconds = 0;
// Query the database using Min Hash
Dictionary<int, QueryStats> allCandidates = QueryFingerprintManager.QueryOneSongMinHash(
signatures,
dbService,
minHash,
lshHashTables,
lshGroupsPerKey,
thresholdTables,
ref elapsedMiliseconds);
IEnumerable<int> ids = allCandidates.Select(p => p.Key);
IList<Track> tracks = dbService.ReadTrackById(ids);
// Order by Hamming Similarity
// TODO: What does the 0.4 number do here?
// there doesn't seem to be any change using another number?!
// Using PLINQ
//OrderedParallelQuery<KeyValuePair<int, QueryStats>> order = allCandidates.AsParallel()
IOrderedEnumerable<KeyValuePair<int, QueryStats>> order = allCandidates
.OrderBy((pair) => pair.Value.OrderingValue =
pair.Value.HammingDistance / pair.Value.NumberOfTotalTableVotes
+ 0.4 * pair.Value.MinHammingDistance);
// Join on the ID properties.
var fingerprintList = (from o in order
join track in tracks on o.Key equals track.Id
select new FindSimilar.QueryResult {
Id = track.Id,
Path = track.FilePath,
Duration = track.TrackLengthMs,
Similarity = o.Value.Similarity
}).ToList();
return fingerprintList;
}
/// <summary>
/// Add the log spectrogram matrix as a Statistical Cluster Model Similarity class to the database
/// </summary>
/// <param name="logSpectrogramMatrix">log spectrogram matrix</param>
/// <param name="fileName">clean filename without extension</param>
/// <param name="fullFilePath">full file path</param>
/// <param name="duration">duration in ms</param>
/// <param name="db">database</param>
/// <param name="trackId">track id to insert</param>
/// <param name="doOutputDebugInfo">decide whether to output debug info like spectrogram and audiofile (default value can be set)</param>
/// <param name="useHaarWavelet">decide whether to use haar wavelet compression or DCT compression</param>
/// <returns>true if successful</returns>
private static bool AnalyseAndAddScmsUsingLogSpectrogram(Comirva.Audio.Util.Maths.Matrix logSpectrogramMatrix,
WorkUnitParameterObject param,
Db db,
int trackId,
bool doOutputDebugInfo=DEFAULT_DEBUG_INFO,
bool useHaarWavelet = true)
{
// Insert Statistical Cluster Model Similarity Audio Feature
string fileName = param.FileName;
Comirva.Audio.Util.Maths.Matrix scmsMatrix = null;
if (useHaarWavelet) {
#region Wavelet Transform
int lastHeight = 0;
int lastWidth = 0;
scmsMatrix = mfccMirage.ApplyWaveletCompression(ref logSpectrogramMatrix, out lastHeight, out lastWidth);
#if DEBUG
if (DEBUG_INFO_VERBOSE) {
if (DEBUG_OUTPUT_TEXT) scmsMatrix.WriteAscii(fileName + "_waveletdata.ascii");
}
#endif
if (doOutputDebugInfo) {
scmsMatrix.DrawMatrixImageLogValues(fileName + "_waveletdata.png", true);
}
#if DEBUG
if (DEBUG_DO_INVERSE_TESTS) {
#region Inverse Wavelet
// try to do an inverse wavelet transform
Comirva.Audio.Util.Maths.Matrix stftdata_inverse_wavelet = mfccMirage.InverseWaveletCompression(ref scmsMatrix, lastHeight, lastWidth, logSpectrogramMatrix.Rows, logSpectrogramMatrix.Columns);
if (DEBUG_OUTPUT_TEXT) stftdata_inverse_wavelet.WriteCSV(fileName + "_specgramlog_inverse_wavelet.csv", ";");
stftdata_inverse_wavelet.DrawMatrixImageLogValues(fileName + "_specgramlog_inverse_wavelet.png", true);
#endregion
}
#endif
#endregion
} else {
#region DCT Transform
// It seems the Mirage way of applying the DCT is slightly faster than the
// Comirva way due to less loops
scmsMatrix = mfccMirage.ApplyDCT(ref logSpectrogramMatrix);
#if DEBUG
if (DEBUG_INFO_VERBOSE) {
if (DEBUG_OUTPUT_TEXT) scmsMatrix.WriteAscii(fileName + "_mfccdata.ascii");
}
#endif
if (doOutputDebugInfo) {
scmsMatrix.DrawMatrixImageLogValues(fileName + "_mfccdata.png", true);
}
#if DEBUG
if (DEBUG_DO_INVERSE_TESTS) {
#region Inverse MFCC
// try to do an inverse mfcc
Comirva.Audio.Util.Maths.Matrix stftdata_inverse_mfcc = mfccMirage.InverseDCT(ref scmsMatrix);
if (DEBUG_OUTPUT_TEXT) stftdata_inverse_mfcc.WriteCSV(fileName + "_stftdata_inverse_mfcc.csv", ";");
stftdata_inverse_mfcc.DrawMatrixImageLogValues(fileName + "_specgramlog_inverse_mfcc.png", true);
#endregion
}
#endif
#endregion
}
// Store in a Statistical Cluster Model Similarity class.
// i.e. a Gaussian representation of a song
Scms audioFeature = Scms.GetScms(scmsMatrix, fileName);
if (audioFeature != null) {
// Store image if debugging
if (doOutputDebugInfo) {
audioFeature.Image = scmsMatrix.DrawMatrixImageLogValues(fileName + "_featuredata.png", true, false, 0, 0, true);
}
// Store bitstring hash as well
string hashString = GetBitString(scmsMatrix);
audioFeature.BitString = hashString;
// Store duration
audioFeature.Duration = (long) param.DurationInMs;
// Store file name
audioFeature.Name = param.PathToAudioFile;
// Add to database
int id = trackId;
if (db.AddTrack(ref id, audioFeature) == -1) {
Console.Out.WriteLine("Failed! Could not add audio feature to database ({0})!", fileName);
return false;
} else {
return true;
}
} else {
Console.Out.WriteLine("Error! Could not compute the Scms for '{0}'!", fileName);
return false;
}
}
//private static Mfcc mfccOptimized = new Mfcc(WINDOW_SIZE, SAMPLING_RATE, MEL_COEFFICIENTS, MFCC_COEFFICIENTS);
//private static MFCC mfccComirva = new MFCC(SAMPLING_RATE, WINDOW_SIZE, MFCC_COEFFICIENTS, true, 20.0, SAMPLING_RATE/2, MEL_COEFFICIENTS);
#endif
#region Methods
public static bool AnalyzeAndAdd(FileInfo filePath, Db db, DatabaseService databaseService, bool doOutputDebugInfo=DEFAULT_DEBUG_INFO, bool useHaarWavelet = true)
{
DbgTimer t = new DbgTimer();
t.Start ();
float[] audiodata = AudioFileReader.Decode(filePath.FullName, SAMPLING_RATE, SECONDS_TO_ANALYZE);
if (audiodata == null || audiodata.Length == 0) {
Dbg.WriteLine("Error! - No Audio Found");
return false;
}
// Read TAGs using BASS
FindSimilar.AudioProxies.BassProxy bass = FindSimilar.AudioProxies.BassProxy.Instance;
Un4seen.Bass.AddOn.Tags.TAG_INFO tag_info = bass.GetTagInfoFromFile(filePath.FullName);
// Name of file being processed
string name = StringUtils.RemoveNonAsciiCharacters(Path.GetFileNameWithoutExtension(filePath.Name));
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE) {
if (DEBUG_OUTPUT_TEXT) WriteAscii(audiodata, name + "_audiodata.ascii");
if (DEBUG_OUTPUT_TEXT) WriteF3Formatted(audiodata, name + "_audiodata.txt");
}
#endif
if (doOutputDebugInfo) {
DrawGraph(MathUtils.FloatToDouble(audiodata), name + "_audiodata.png");
}
// Calculate duration in ms
double duration = (double) audiodata.Length / SAMPLING_RATE * 1000;
// Explode samples to the range of 16 bit shorts (–32,768 to 32,767)
// Matlab multiplies with 2^15 (32768)
// e.g. if( max(abs(speech))<=1 ), speech = speech * 2^15; end;
MathUtils.Multiply(ref audiodata, AUDIO_MULTIPLIER); // 65536
// zero pad if the audio file is too short to perform a mfcc
if (audiodata.Length < (fingerprintingConfig.WdftSize + fingerprintingConfig.Overlap))
{
int lenNew = fingerprintingConfig.WdftSize + fingerprintingConfig.Overlap;
Array.Resize<float>(ref audiodata, lenNew);
}
// Get fingerprint signatures using the Soundfingerprinting methods
IPermutations permutations = new LocalPermutations("Soundfingerprinting\\perms.csv", ",");
Repository repository = new Repository(permutations, databaseService, fingerprintService);
// Image Service
ImageService imageService = new ImageService(
fingerprintService.SpectrumService,
fingerprintService.WaveletService);
// work config
WorkUnitParameterObject param = new WorkUnitParameterObject();
param.FingerprintingConfiguration = fingerprintingConfig;
param.AudioSamples = audiodata;
param.PathToAudioFile = filePath.FullName;
param.MillisecondsToProcess = SECONDS_TO_ANALYZE * 1000;
param.StartAtMilliseconds = 0;
// build track
Track track = new Track();
track.Title = name;
track.TrackLengthMs = (int) duration;
track.FilePath = filePath.FullName;
track.Id = -1; // this will be set by the insert method
#region parse tag_info
if (tag_info != null) {
Dictionary<string, string> tags = new Dictionary<string, string>();
//if (tag_info.title != string.Empty) tags.Add("title", tag_info.title);
if (tag_info.artist != string.Empty) tags.Add("artist", tag_info.artist);
if (tag_info.album != string.Empty) tags.Add("album", tag_info.album);
if (tag_info.albumartist != string.Empty) tags.Add("albumartist", tag_info.albumartist);
if (tag_info.year != string.Empty) tags.Add("year", tag_info.year);
if (tag_info.comment != string.Empty) tags.Add("comment", tag_info.comment);
if (tag_info.genre != string.Empty) tags.Add("genre", tag_info.genre);
if (tag_info.track != string.Empty) tags.Add("track", tag_info.track);
if (tag_info.disc != string.Empty) tags.Add("disc", tag_info.disc);
if (tag_info.copyright != string.Empty) tags.Add("copyright", tag_info.copyright);
if (tag_info.encodedby != string.Empty) tags.Add("encodedby", tag_info.encodedby);
if (tag_info.composer != string.Empty) tags.Add("composer", tag_info.composer);
if (tag_info.publisher != string.Empty) tags.Add("publisher", tag_info.publisher);
if (tag_info.lyricist != string.Empty) tags.Add("lyricist", tag_info.lyricist);
if (tag_info.remixer != string.Empty) tags.Add("remixer", tag_info.remixer);
if (tag_info.producer != string.Empty) tags.Add("producer", tag_info.producer);
if (tag_info.bpm != string.Empty) tags.Add("bpm", tag_info.bpm);
//if (tag_info.filename != string.Empty) tags.Add("filename", tag_info.filename);
tags.Add("channelinfo", tag_info.channelinfo.ToString());
//if (tag_info.duration > 0) tags.Add("duration", tag_info.duration.ToString());
if (tag_info.bitrate > 0) tags.Add("bitrate", tag_info.bitrate.ToString());
if (tag_info.replaygain_track_gain != -100f) tags.Add("replaygain_track_gain", tag_info.replaygain_track_gain.ToString());
if (tag_info.replaygain_track_peak != -1f) tags.Add("replaygain_track_peak", tag_info.replaygain_track_peak.ToString());
if (tag_info.conductor != string.Empty) tags.Add("conductor", tag_info.conductor);
if (tag_info.grouping != string.Empty) tags.Add("grouping", tag_info.grouping);
if (tag_info.mood != string.Empty) tags.Add("mood", tag_info.mood);
if (tag_info.rating != string.Empty) tags.Add("rating", tag_info.rating);
if (tag_info.isrc != string.Empty) tags.Add("isrc", tag_info.isrc);
foreach(var nativeTag in tag_info.NativeTags) {
string[] keyvalue = nativeTag.Split('=');
tags.Add(keyvalue[0], keyvalue[1]);
}
track.Tags = tags;
}
#endregion
double[][] logSpectrogram;
if (repository.InsertTrackInDatabaseUsingSamples(track, 25, 4, param, out logSpectrogram)) {
// store logSpectrogram as Matrix
Comirva.Audio.Util.Maths.Matrix logSpectrogramMatrix = new Comirva.Audio.Util.Maths.Matrix(logSpectrogram);
logSpectrogramMatrix = logSpectrogramMatrix.Transpose();
#region Debug for Soundfingerprinting Method
if (doOutputDebugInfo) {
imageService.GetLogSpectralImages(logSpectrogram, fingerprintingConfig.Stride, fingerprintingConfig.FingerprintLength, fingerprintingConfig.Overlap, 2).Save(name + "_specgram_logimages.png");
logSpectrogramMatrix.DrawMatrixImageLogValues(name + "_specgram_logimage.png", true);
if (DEBUG_OUTPUT_TEXT) {
logSpectrogramMatrix.WriteCSV(name + "_specgram_log.csv", ";");
}
}
#endregion
#region Insert Statistical Cluster Model Similarity Audio Feature as well
Comirva.Audio.Util.Maths.Matrix scmsMatrix = null;
if (useHaarWavelet) {
#region Wavelet Transform
int lastHeight = 0;
int lastWidth = 0;
scmsMatrix = mfccMirage.ApplyWaveletCompression(ref logSpectrogramMatrix, out lastHeight, out lastWidth);
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE) {
if (DEBUG_OUTPUT_TEXT) scmsMatrix.WriteAscii(name + "_waveletdata.ascii");
}
#endif
if (doOutputDebugInfo) {
scmsMatrix.DrawMatrixImageLogValues(name + "_waveletdata.png", true);
}
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE) {
#region Inverse Wavelet
// try to do an inverse wavelet transform
Comirva.Audio.Util.Maths.Matrix stftdata_inverse_wavelet = mfccMirage.InverseWaveletCompression(ref scmsMatrix, lastHeight, lastWidth, logSpectrogramMatrix.Rows, logSpectrogramMatrix.Columns);
if (DEBUG_OUTPUT_TEXT) stftdata_inverse_wavelet.WriteCSV(name + "_specgramlog_inverse_wavelet.csv", ";");
stftdata_inverse_wavelet.DrawMatrixImageLogValues(name + "_specgramlog_inverse_wavelet.png", true);
#endregion
}
#endif
#endregion
} else {
#region DCT Transform
// It seems the Mirage way of applying the DCT is slightly faster than the
// Comirva way due to less loops
scmsMatrix = mfccMirage.ApplyDCT(ref logSpectrogramMatrix);
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE) {
if (DEBUG_OUTPUT_TEXT) scmsMatrix.WriteAscii(name + "_mfccdata.ascii");
}
#endif
if (doOutputDebugInfo) {
scmsMatrix.DrawMatrixImageLogValues(name + "_mfccdata.png", true);
}
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE) {
#region Inverse MFCC
// try to do an inverse mfcc
Comirva.Audio.Util.Maths.Matrix stftdata_inverse_mfcc = mfccMirage.InverseDCT(ref scmsMatrix);
if (DEBUG_OUTPUT_TEXT) stftdata_inverse_mfcc.WriteCSV(name + "_stftdata_inverse_mfcc.csv", ";");
stftdata_inverse_mfcc.DrawMatrixImageLogValues(name + "_specgramlog_inverse_mfcc.png", true);
#endregion
}
#endif
#endregion
}
// Store in a Statistical Cluster Model Similarity class.
// A Gaussian representation of a song
Scms audioFeature = Scms.GetScms(scmsMatrix, name);
if (audioFeature != null) {
// Store image if debugging
if (doOutputDebugInfo) {
audioFeature.Image = scmsMatrix.DrawMatrixImageLogValues(name + "_featuredata.png", true, false, 0, 0, true);
}
// Store bitstring hash as well
string hashString = GetBitString(scmsMatrix);
audioFeature.BitString = hashString;
// Store duration
audioFeature.Duration = (long) duration;
// Store file name
audioFeature.Name = filePath.FullName;
int id = track.Id;
if (db.AddTrack(ref id, audioFeature) == -1) {
Console.Out.WriteLine("Failed! Could not add audioFeature to database {0}!", name);
}
}
#endregion
} else {
// failed
return false;
}
Dbg.WriteLine ("AnalyzeAndAdd - Total Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return true;
}
private static List<bool[]> GetFingerprintSignatures(FingerprintService fingerprintService, float[] samples, string name)
{
Mirage.DbgTimer t = new Mirage.DbgTimer();
t.Start();
// work config
WorkUnitParameterObject param = new WorkUnitParameterObject();
param.FingerprintingConfiguration = fingerprintingConfigCreation;
// Get fingerprints
double[][] LogSpectrogram;
List<double[][]> spectralImages;
List<bool[]> fingerprints = fingerprintService.CreateFingerprintsFromAudioSamples(samples, param, out LogSpectrogram, out spectralImages);
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE) {
// Image Service
ImageService imageService =
new ImageService(fingerprintService.SpectrumService, fingerprintService.WaveletService);
int width = param.FingerprintingConfiguration.FingerprintLength;
int height = param.FingerprintingConfiguration.LogBins;
imageService.GetImageForFingerprints(fingerprints, width, height, 2).Save(name + "_fingerprints.png");
}
#endif
Mirage.Dbg.WriteLine("GetFingerprintSignatures Execution Time: " + t.Stop().TotalMilliseconds + " ms");
return fingerprints;
}
public static AudioFeature AnalyzeSoundfingerprinting(FileInfo filePath, bool doOutputDebugInfo=DEFAULT_DEBUG_INFO, bool useHaarWavelet = true)
{
DbgTimer t = new DbgTimer();
t.Start ();
float[] audiodata = AudioFileReader.Decode(filePath.FullName, SAMPLING_RATE, SECONDS_TO_ANALYZE);
if (audiodata == null || audiodata.Length == 0) {
Dbg.WriteLine("Error! - No Audio Found");
return null;
}
// Read TAGs using BASS
FindSimilar.AudioProxies.BassProxy bass = FindSimilar.AudioProxies.BassProxy.Instance;
Un4seen.Bass.AddOn.Tags.TAG_INFO tag_info = bass.GetTagInfoFromFile(filePath.FullName);
// Name of file being processed
string name = StringUtils.RemoveNonAsciiCharacters(Path.GetFileNameWithoutExtension(filePath.Name));
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE) {
if (DEBUG_OUTPUT_TEXT) WriteAscii(audiodata, name + "_audiodata.ascii");
if (DEBUG_OUTPUT_TEXT) WriteF3Formatted(audiodata, name + "_audiodata.txt");
}
#endif
if (doOutputDebugInfo) {
DrawGraph(MathUtils.FloatToDouble(audiodata), name + "_audiodata.png");
}
// Calculate duration in ms
double duration = (double) audiodata.Length / SAMPLING_RATE * 1000;
// zero pad if the audio file is too short to perform a mfcc
if (audiodata.Length < (fingerprintingConfig.WdftSize + fingerprintingConfig.Overlap))
{
int lenNew = fingerprintingConfig.WdftSize + fingerprintingConfig.Overlap;
Array.Resize<float>(ref audiodata, lenNew);
}
// Get fingerprint signatures using the Soundfingerprinting methods
// Get database
DatabaseService databaseService = DatabaseService.Instance;
IPermutations permutations = new LocalPermutations("Soundfingerprinting\\perms.csv", ",");
Repository repository = new Repository(permutations, databaseService, fingerprintService);
// Image Service
ImageService imageService = new ImageService(
fingerprintService.SpectrumService,
fingerprintService.WaveletService);
// work config
WorkUnitParameterObject param = new WorkUnitParameterObject();
param.FingerprintingConfiguration = fingerprintingConfig;
param.AudioSamples = audiodata;
param.PathToAudioFile = filePath.FullName;
param.MillisecondsToProcess = SECONDS_TO_ANALYZE * 1000;
param.StartAtMilliseconds = 0;
// build track
Track track = new Track();
track.Title = name;
track.TrackLengthMs = (int) duration;
track.FilePath = filePath.FullName;
track.Id = -1; // this will be set by the insert method
#region parse tag_info
if (tag_info != null) {
Dictionary<string, string> tags = new Dictionary<string, string>();
//if (tag_info.title != string.Empty) tags.Add("title", tag_info.title);
if (tag_info.artist != string.Empty) tags.Add("artist", tag_info.artist);
if (tag_info.album != string.Empty) tags.Add("album", tag_info.album);
if (tag_info.albumartist != string.Empty) tags.Add("albumartist", tag_info.albumartist);
if (tag_info.year != string.Empty) tags.Add("year", tag_info.year);
if (tag_info.comment != string.Empty) tags.Add("comment", tag_info.comment);
if (tag_info.genre != string.Empty) tags.Add("genre", tag_info.genre);
if (tag_info.track != string.Empty) tags.Add("track", tag_info.track);
if (tag_info.disc != string.Empty) tags.Add("disc", tag_info.disc);
if (tag_info.copyright != string.Empty) tags.Add("copyright", tag_info.copyright);
if (tag_info.encodedby != string.Empty) tags.Add("encodedby", tag_info.encodedby);
if (tag_info.composer != string.Empty) tags.Add("composer", tag_info.composer);
if (tag_info.publisher != string.Empty) tags.Add("publisher", tag_info.publisher);
if (tag_info.lyricist != string.Empty) tags.Add("lyricist", tag_info.lyricist);
if (tag_info.remixer != string.Empty) tags.Add("remixer", tag_info.remixer);
if (tag_info.producer != string.Empty) tags.Add("producer", tag_info.producer);
if (tag_info.bpm != string.Empty) tags.Add("bpm", tag_info.bpm);
//if (tag_info.filename != string.Empty) tags.Add("filename", tag_info.filename);
tags.Add("channelinfo", tag_info.channelinfo.ToString());
//if (tag_info.duration > 0) tags.Add("duration", tag_info.duration.ToString());
if (tag_info.bitrate > 0) tags.Add("bitrate", tag_info.bitrate.ToString());
if (tag_info.replaygain_track_gain != -100f) tags.Add("replaygain_track_gain", tag_info.replaygain_track_gain.ToString());
if (tag_info.replaygain_track_peak != -1f) tags.Add("replaygain_track_peak", tag_info.replaygain_track_peak.ToString());
if (tag_info.conductor != string.Empty) tags.Add("conductor", tag_info.conductor);
if (tag_info.grouping != string.Empty) tags.Add("grouping", tag_info.grouping);
if (tag_info.mood != string.Empty) tags.Add("mood", tag_info.mood);
if (tag_info.rating != string.Empty) tags.Add("rating", tag_info.rating);
if (tag_info.isrc != string.Empty) tags.Add("isrc", tag_info.isrc);
foreach(var nativeTag in tag_info.NativeTags) {
string[] keyvalue = nativeTag.Split('=');
tags.Add(keyvalue[0], keyvalue[1]);
}
track.Tags = tags;
}
#endregion
AudioFeature audioFeature = null;
double[][] logSpectrogram;
if (repository.InsertTrackInDatabaseUsingSamples(track, 25, 4, param, out logSpectrogram)) {
if (doOutputDebugInfo) {
imageService.GetLogSpectralImages(logSpectrogram, fingerprintingConfig.Stride, fingerprintingConfig.FingerprintLength, fingerprintingConfig.Overlap, 2).Save(name + "_specgram_logimages.png");
Comirva.Audio.Util.Maths.Matrix logSpectrogramMatrix = new Comirva.Audio.Util.Maths.Matrix(logSpectrogram);
logSpectrogramMatrix = logSpectrogramMatrix.Transpose();
logSpectrogramMatrix.DrawMatrixImageLogValues(name + "_specgram_logimage.png", true);
if (DEBUG_OUTPUT_TEXT) {
logSpectrogramMatrix.WriteCSV(name + "_specgram_log.csv", ";");
}
}
audioFeature = new DummyAudioFeature();
// Store duration
audioFeature.Duration = (long) duration;
// Store file name
audioFeature.Name = filePath.FullName;
} else {
// failed
}
Dbg.WriteLine ("Soundfingerprinting - Total Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return audioFeature;
}
private static bool AnalyseAndAddScmsUsingFingerprints(List<bool[]> fingerprints,
WorkUnitParameterObject param,
Db db,
int trackId,
bool doOutputDebugInfo=DEFAULT_DEBUG_INFO)
{
DbgTimer t = new DbgTimer();
t.Start ();
// Insert Statistical Cluster Model Similarity Audio Feature
string fileName = param.FileName;
int fingerprintWidth = param.FingerprintingConfiguration.FingerprintLength;
int fingerprintHeight = param.FingerprintingConfiguration.LogBins;
int fingerprintCount = 0;
foreach (bool[] fingerprint in fingerprints) {
fingerprintCount++;
Comirva.Audio.Util.Maths.Matrix scmsMatrix = new Comirva.Audio.Util.Maths.Matrix(fingerprintWidth, fingerprintHeight);
for (int i = 0; i < fingerprintWidth /*128*/; i++) {
for (int j = 0; j < fingerprintHeight /*32*/; j++) {
// Negative Numbers = 01
// Positive Numbers = 10
// Zeros = 00
bool v1 = fingerprint[(2 * fingerprintHeight * i) + (2 * j)];
bool v2 = fingerprint[(2 * fingerprintHeight * i) + (2 * j) + 1];
if (v1) {
scmsMatrix.MatrixData[i][j] = 2.0;
} else if (v2) {
scmsMatrix.MatrixData[i][j] = 0.0;
} else {
scmsMatrix.MatrixData[i][j] = 1.0;
}
}
}
if (doOutputDebugInfo) {
scmsMatrix.DrawMatrixImage(String.Format("{0}_fingerprint_{1}.png", fileName, fingerprintCount), fingerprintWidth, fingerprintHeight);
}
#region Store in a Statistical Cluster Model Similarity class.
Scms audioFeature = Scms.GetScmsNoInverse(scmsMatrix, fileName);
if (audioFeature != null) {
// Store bitstring hash as well
audioFeature.BitString = GetBitString(fingerprint);
// Store duration
audioFeature.Duration = (long) param.DurationInMs;
// Store file name
audioFeature.Name = param.PathToAudioFile;
// Add to database
int id = trackId;
if (db.AddTrack(audioFeature) == -1) {
Console.Out.WriteLine("Failed! Could not add audio feature to database ({0})!", fileName);
return false;
}
} else {
return false;
}
#endregion
}
Dbg.WriteLine ("AnalyseAndAddScmsUsingFingerprints - Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return true;
}
/// <summary>
/// Find Similar Tracks using passed audio samples as input and return a Dictionary
/// </summary>
/// <param name="lshHashTables">Number of hash tables from the database</param>
/// <param name="lshGroupsPerKey">Number of groups per hash table</param>
/// <param name="thresholdTables">Minimum number of hash tables that must be found for one signature to be considered a candidate (0 and 1 = return all candidates, 2+ = return only exact matches)</param>
/// <param name="param">Audio File Work Unit Parameter Object</param>
/// <returns>a dictionary of perceptually similar tracks</returns>
public Dictionary<Track, double> FindSimilarFromAudioSamples(
int lshHashTables,
int lshGroupsPerKey,
int thresholdTables,
WorkUnitParameterObject param) {
DbgTimer t = new DbgTimer();
t.Start ();
// Get fingerprints
double[][] logSpectrogram;
List<bool[]> signatures = fingerprintService.CreateFingerprintsFromAudioSamples(param.AudioSamples, param, out logSpectrogram);
long elapsedMiliseconds = 0;
// Query the database using Min Hash
Dictionary<int, QueryStats> allCandidates = QueryFingerprintManager.QueryOneSongMinHash(
signatures,
dbService,
minHash,
lshHashTables,
lshGroupsPerKey,
thresholdTables,
ref elapsedMiliseconds);
IEnumerable<int> ids = allCandidates.Select(p => p.Key);
IList<Track> tracks = dbService.ReadTrackById(ids);
// Order by Hamming Similarity
// Using PLINQ
//OrderedParallelQuery<KeyValuePair<int, QueryStats>> order = allCandidates.AsParallel()
IOrderedEnumerable<KeyValuePair<int, QueryStats>> order = allCandidates
.OrderBy((pair) => pair.Value.OrderingValue =
pair.Value.HammingDistance / pair.Value.NumberOfTotalTableVotes
+ 0.4 * pair.Value.MinHammingDistance);
// Join on the ID properties.
var joined = from o in order
join track in tracks on o.Key equals track.Id
select new { track, o.Value.Similarity };
Dictionary<Track, double> stats = joined.ToDictionary(Key => Key.track, Value => Value.Similarity);
Dbg.WriteLine ("Find Similar From Audio Samples - Total Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return stats;
}
