public List <bool[]> CreateFingerprintsFromAudioSamples(float[] samples, WorkUnitParameterObject param, out double[][] logSpectrogram)
{
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));
}
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));
}
/// <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);
}
/// <summary>
/// Find Similar Tracks using passed audio file 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>
/// <param name="splashScreen">The "please wait" splash screen (or null)</param>
/// <returns>a dictionary of perceptually similar tracks</returns>
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);
}
public WorkUnit(WorkUnitParameterObject workUnitParameterObject)
{
this.workUnitParameterObject = workUnitParameterObject;
}
/// <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>
/// <param name="logSpectrogram">Return the log spectrogram in an out parameter</param>
/// <param name="fingerprints">Return the fingerprints in an out parameter</param>
/// <returns></returns>
public bool InsertTrackInDatabaseUsingSamples(Track track, int hashTables, int hashKeys, WorkUnitParameterObject param, out double[][] logSpectrogram, out List <bool[]> fingerprints)
{
if (dbService.InsertTrack(track))
{
// return both logSpectrogram and fingerprints in the out variables
fingerprints = fingerprintService.CreateFingerprintsFromAudioSamples(param.AudioSamples, param, out logSpectrogram);
List <Fingerprint> inserted = AssociateFingerprintsToTrack(fingerprints, track.Id);
if (dbService.InsertFingerprint(inserted))
{
return(HashFingerprintsUsingMinHash(inserted, track, hashTables, hashKeys));
}
else
{
return(false);
}
}
else
{
logSpectrogram = null;
fingerprints = null;
return(false);
}
}
/// <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);
}
public CompareAudioForm()
{
//
// The InitializeComponent() call is required for Windows Forms designer support.
//
InitializeComponent();
//
// TODO: Add constructor code after the InitializeComponent() call.
//
// Instansiate Soundfingerprinting Repository
FingerprintService fingerprintService = Analyzer.GetSoundfingerprintingService();
this.databaseService = DatabaseService.Instance;
IPermutations permutations = new LocalPermutations("Soundfingerprinting\\perms.csv", ",");
//IPermutations permutations = new LocalPermutations("Soundfingerprinting\\perms-new.csv", ",");
IFingerprintingConfiguration fingerprintingConfigCreation = new FullFrequencyFingerprintingConfiguration();
repository = new Repository(permutations, databaseService, fingerprintService);
ImageService imageService = new ImageService(fingerprintService.SpectrumService, fingerprintService.WaveletService);
FileInfo filePathAudio1 = new FileInfo(@"C:\Users\perivar.nerseth\Music\Test Samples Database\VDUB1 Snare 004.wav");
FileInfo filePathAudio2 = new FileInfo(@"C:\Users\perivar.nerseth\Music\Test Samples Search\VDUB1 Snare 004 - Start.wav");
int fingerprintsPerRow = 2;
double[][] logSpectrogram1 = null;
double[][] logSpectrogram2 = null;
List <bool[]> fingerprints1 = null;
List <bool[]> fingerprints2 = null;
WorkUnitParameterObject file1Param = Analyzer.GetWorkUnitParameterObjectFromAudioFile(filePathAudio1);
if (file1Param != null)
{
file1Param.FingerprintingConfiguration = fingerprintingConfigCreation;
// Get fingerprints
fingerprints1 = fingerprintService.CreateFingerprintsFromAudioSamples(file1Param.AudioSamples, file1Param, out logSpectrogram1);
pictureBox1.Image = imageService.GetSpectrogramImage(logSpectrogram1, logSpectrogram1.Length, logSpectrogram1[0].Length);
pictureBoxWithInterpolationMode1.Image = imageService.GetImageForFingerprints(fingerprints1, file1Param.FingerprintingConfiguration.FingerprintLength, file1Param.FingerprintingConfiguration.LogBins, fingerprintsPerRow);
}
WorkUnitParameterObject file2Param = Analyzer.GetWorkUnitParameterObjectFromAudioFile(filePathAudio2);
if (file2Param != null)
{
file2Param.FingerprintingConfiguration = fingerprintingConfigCreation;
// Get fingerprints
fingerprints2 = fingerprintService.CreateFingerprintsFromAudioSamples(file2Param.AudioSamples, file2Param, out logSpectrogram2);
pictureBox2.Image = imageService.GetSpectrogramImage(logSpectrogram2, logSpectrogram2.Length, logSpectrogram2[0].Length);
pictureBoxWithInterpolationMode2.Image = imageService.GetImageForFingerprints(fingerprints2, file2Param.FingerprintingConfiguration.FingerprintLength, file2Param.FingerprintingConfiguration.LogBins, fingerprintsPerRow);
}
MinHash minHash = repository.MinHash;
// only use the first signatures
bool[] signature1 = fingerprints1[0];
bool[] signature2 = fingerprints2[0];
if (signature1 != null && signature2 != null)
{
int hammingDistance = MinHash.CalculateHammingDistance(signature1, signature2);
double jaqSimilarity = MinHash.CalculateJaqSimilarity(signature1, signature2);
lblSimilarity.Text = String.Format("Hamming: {0} JAQ: {1}", hammingDistance, jaqSimilarity);
}
}
