/// <summary>
/// Query one specific song using MinHash algorithm.
/// </summary>
/// <param name="signatures">Signature signatures from a song</param>
/// <param name="dbService">DatabaseService used to query the underlying database</param>
/// <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 = return all candidates, 2+ = return only exact matches)</param>
/// <param name="queryTime">Set by the method, representing the query length</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>Dictionary with Tracks ID's and the Query Statistics</returns>
public static Dictionary<Int32, QueryStats> QueryOneSongMinHash(
IEnumerable<bool[]> signatures,
DatabaseService dbService,
MinHash minHash,
int lshHashTables,
int lshGroupsPerKey,
int thresholdTables,
ref long queryTime,
bool doSearchEverything = false,
SplashSceenWaitingForm splashScreen = null)
{
Stopwatch stopWatch = new Stopwatch();
stopWatch.Start();
int signatureCounter = 0;
int signatureTotalCount = signatures.Count();
Dictionary<int, QueryStats> stats = new Dictionary<int, QueryStats>();
foreach (bool[] signature in signatures) {
#region Please Wait Splash Screen Cancel Event
// check if the user clicked cancel
if (splashScreen.CancellationPending) {
break;
}
#endregion
if (signature == null) {
continue;
}
IDictionary<int, IList<HashBinMinHash>> candidates = null;
if (doSearchEverything) {
candidates = dbService.ReadAllFingerprints();
} else {
// Compute Min Hash on randomly selected fingerprint
int[] bin = minHash.ComputeMinHashSignature(signature);
// Find all hashbuckets to care about
Dictionary<int, long> hashes = minHash.GroupMinHashToLSHBuckets(bin, lshHashTables, lshGroupsPerKey);
long[] hashbuckets = hashes.Values.ToArray();
// Find all candidates by querying the database for those hashbuckets
candidates = dbService.ReadFingerprintsByHashBucketLsh(hashbuckets);
}
// Reduce the potential candidates list if the number of hash tables found for each signature are less than the threshold
Dictionary<int, IList<HashBinMinHash>> potentialCandidates = SelectPotentialMatchesOutOfEntireDataset(candidates, thresholdTables);
// get the final candidate list by only using the potential candidate list
if (potentialCandidates.Count > 0) {
IList<Fingerprint> fingerprints = dbService.ReadFingerprintById(potentialCandidates.Keys);
Dictionary<Fingerprint, int> finalCandidates = fingerprints.ToDictionary(finger => finger, finger => potentialCandidates[finger.Id].Count);
ArrangeCandidatesAccordingToFingerprints(signature,
finalCandidates,
lshHashTables,
lshGroupsPerKey,
stats);
}
#region Please Wait Splash Screen Update
// calculate a percentage between 5 and 90
int percentage = (int) ((float) (signatureCounter) / (float) signatureTotalCount * 85) + 5;
if (splashScreen != null) splashScreen.SetProgress(percentage, String.Format("Searching for similar fingerprints.\n(Signature {0} of {1})", signatureCounter+1, signatureTotalCount));
signatureCounter++;
#endregion Updat
}
stopWatch.Stop();
queryTime = stopWatch.ElapsedMilliseconds; /*Set the query Time parameter*/
return stats;
}
/// <summary>
/// Query the database for perceptually similar tracks using the sound fingerprinting methods
/// </summary>
/// <param name="filePath">input file</param>
/// <param name="repository">the database (repository)</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="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 query results objects (e.g. similar tracks)</returns>
public static List<FindSimilar.QueryResult> SimilarTracksSoundfingerprintingList(FileInfo filePath,
Repository repository,
int thresholdTables,
bool optimizeSignatureCount,
bool doSearchEverything,
SplashSceenWaitingForm splashScreen) {
DbgTimer t = new DbgTimer();
t.Start ();
if (splashScreen != null) splashScreen.SetProgress(0, "Reading audio file ...");
// get work config from the audio file
WorkUnitParameterObject param = GetWorkUnitParameterObjectFromAudioFile(filePath);
if (param == null) {
if (splashScreen != null) splashScreen.SetProgress(0, "Failed reading audio file!");
return null;
}
param.FingerprintingConfiguration = fingerprintingConfigQuerying;
if (splashScreen != null) splashScreen.SetProgress(1, "Successfully reading audio file!");
// This is how the threshold tables work:
// For each signature created from a query file we retrieve a number of candidates
// based on how many fingerprints that are associated to the same hash bucket.
// if the number of fingerprints associated to the same hash bucket is relatively high
// the likelyhood for this being an exact match is also very high.
// Therefore a value of 0 or 1 basically means return every track that has an association
// to the same hash bucket, while a number higher than that increases the accuracy for
// only matching identical matches.
// 0 and 1 returns many matches
// 2 returns sometimes only the one we search for (exact match)
List<FindSimilar.QueryResult> similarFiles = repository.FindSimilarFromAudioSamplesList(param.FingerprintingConfiguration.NumberOfHashTables,
param.FingerprintingConfiguration.NumberOfKeys,
thresholdTables,
param,
optimizeSignatureCount,
doSearchEverything,
splashScreen);
Dbg.WriteLine ("SimilarTracksSoundfingerprintingList - Total Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return similarFiles;
}
/// <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;
}
