Skip to content

Commit

Permalink
Add wingbeat profile to Flying Fox
Browse files Browse the repository at this point in the history 
Issue #238 Add Wingbeat profile to recogniser. It picks up wingbeats but also some human speech.
Also set up to read parameters from the config file.
  • Loading branch information
@towsey @atruskie
towsey authored and atruskie committed Aug 28, 2019
1 parent 2cf3847 commit 14a3214
Show file tree
Hide file tree
Showing 2 changed files with 206 additions and 53 deletions.
51 changes: 24 additions & 27 deletions src/AnalysisConfigFiles/RecognizerConfigFiles/Towsey.PteropusSpecies.yml
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
---
SpeciesName: Pteropus species
SpeciesName: Pteropus sp.
CommonName: Flying Fox
AbbreviatedSpeciesName: PteropusSp

Expand All @@ -15,48 +15,45 @@ AbbreviatedSpeciesName: PteropusSp
# Typically the lowest harmonic will lie between 1200 Hz and 3000 Hz and the higher ones evenly spaced.
# This is the tricky bit due to variability but may work to use spectrum auto-correlation.

# Custom settings

SomeExampleSettingA: 3.0
# Common settings
BgNoiseThreshold: 3.0

# Each of these profiles will be analyzed
Profiles:
# The below settings are a template
Standard: &STANDARD
#Standard: &STANDARD
#EventThreshold: 0.2
Territorial:
# min and max of the freq band to search
MinHz: 100
MaxHz: 7000
MinHz: 800
MaxHz: 8000
MinDuration: 0.15
MaxDuration: 0.5
DecibelThreshold: 9.0
Wingbeats:
MinHz: 200
MaxHz: 2000
# duration of DCT in seconds
DctDuration: 0.15
DctDuration: 1.0
# minimum acceptable value of a DCT coefficient
DctThreshold: 0.6
DctThreshold: 0.5
# ignore oscillation rates below the min & above the max threshold
# A sample of four wingbeats yielded an average of 5.1 +/-0.5 beats/s.
# 95% confidence interval (2 SDs) is therefore approx 4-6 beats/s.
# OSCILLATIONS PER SECOND
MinOcilFreq: 50
MaxOcilFreq: 140
MinOscilFreq: 4
MaxOscilFreq: 6
# Minimum and maximum duration for the length of a true call.
MinDuration: 0.1
MaxDuration: 0.5
MinDuration: 1.0
MaxDuration: 10.0
# Event threshold - use this to determine FP / FN trade-off for events.
EventThreshold: 0.30
EventThreshold: 0.50
#Agonist:
# This notation means the Groote profile has all of the settings that the Standard profile has,
# however, the MinHz and MaxHz properties have been overridden.
#Groote:
# <<: *STANDARD
# MinHz: 4000
# MaxHz: 6000
#FemaleRelease:
# <<: *STANDARD
# DctDuration: 0.3

# Standard settings
DoNoiseReduction: false
#BgNoiseThreshold: 3.0

EventThreshold: 0.2

# Resample rate must be 2 X the desired Nyquist
# ResampleRate: 17640
ResampleRate: 22050

## Specifically for AnalyzeLongRecording
Expand Down
208 changes: 182 additions & 26 deletions src/AnalysisPrograms/Recognizers/PteropusSpecies.cs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,8 @@ namespace AnalysisPrograms.Recognizers
using System.IO;
using System.Linq;
using System.Reflection;
using System.Text;
using Accord;
using Acoustics.Shared;
using Acoustics.Shared.ConfigFile;
using AnalysisBase;
Expand All @@ -51,7 +53,7 @@ internal class PteropusSpecies : RecognizerBase

public override string Description => "[STATUS DESCRIPTION] Detects acoustic events for species of Flying Fox, Pteropus species";

private static readonly ILog Log = LogManager.GetLogger(MethodBase.GetCurrentMethod().DeclaringType);
private static readonly ILog log = LogManager.GetLogger(MethodBase.GetCurrentMethod().DeclaringType);

/// <summary>
/// Summarize your results. This method is invoked exactly once per original file.
Expand All @@ -78,11 +80,81 @@ public override void SummariseResults(
/// <param name="outputDirectory">where the recogniser results can be found.</param>
/// <param name="imageWidth"> assuming ????.</param>
/// <returns>recogniser results.</returns>
public override RecognizerResults Recognize(AudioRecording audioRecording, Config configuration, TimeSpan segmentStartOffset, Lazy<IndexCalculateResult[]> getSpectralIndexes, DirectoryInfo outputDirectory, int? imageWidth)
public override RecognizerResults Recognize(AudioRecording audioRecording, Config genericConfig, TimeSpan segmentStartOffset, Lazy<IndexCalculateResult[]> getSpectralIndexes, DirectoryInfo outputDirectory, int? imageWidth)
{
RecognizerResults results = Gruntwork(audioRecording, configuration, outputDirectory, segmentStartOffset);
// Examples of the APIs available. You don't need all of these commands! Pick and choose.
if (ConfigFile.HasProfiles(genericConfig))
{
string[] profileNames = ConfigFile.GetProfileNames<Config>(genericConfig);
int count = profileNames.Length;
var message = new StringBuilder("Found " + count + " config profile(s): ");
foreach (string s in profileNames)
{
message.Append(s + ", ");
}

log.Info(message.ToString());
}
else
{
log.Warn("No configuration profiles found. Three profiles expected for the Flying Fox recogniser.");
}

return results;
RecognizerResults results1 = null;

if (ConfigFile.TryGetProfile(genericConfig, "Territorial", out var profile1))
{
results1 = TerritorialCall(audioRecording, genericConfig, profile1, outputDirectory, segmentStartOffset);
log.Info("Territory event count = " + results1.Events.Count);
}
else
{
log.Warn("Could not access Territorial configuration parameters");
}

RecognizerResults results2 = null;
if (ConfigFile.TryGetProfile(genericConfig, "Wingbeats", out var profile2))
{
results2 = WingBeats(audioRecording, genericConfig, profile2, outputDirectory, segmentStartOffset);
log.Info("Wingbeat event count = " + results2.Events.Count);
}
else
{
log.Warn("Could not access Wingbeats configuration parameters");
}

//RecognizerResults combinedResults = null;

//// combine the results
//return new RecognizerResults()
//{
// Events = results1.Events.AddRange(results2.Events),
// Hits = null,
// ScoreTrack = null,
// Plots = plots,
// Sonogram = sonogram,
//};

// combine the results
if (results1 != null && results2 != null)
{
results1.Events.AddRange(results2.Events);
results1.Plots.AddRange(results2.Plots);
}

//Set following true if you want special debug spectrogram, i.e. with special plots
//In addition, standard spectrograms are produced when you set true in the config file, Towsey.PteropusSpecies.yml.
if (false)
{
//var image = sonogram.GetImageFullyAnnotated("Test");
string speciesName = genericConfig[AnalysisKeys.SpeciesName] ?? "Pteropus species";
var image = SpectrogramTools.GetSonogramPlusCharts(results1.Sonogram, results1.Events, results1.Plots, null);
var opPath = outputDirectory.Combine(FilenameHelpers.AnalysisResultName(Path.GetFileNameWithoutExtension(audioRecording.BaseName), speciesName, "png", "DebugSpectrogram"));
string imageFilename = audioRecording.BaseName + ".profile.png";
image.Save(Path.Combine(outputDirectory.FullName, imageFilename));
}

return results1;
}

/// <summary>
Expand All @@ -93,21 +165,22 @@ public override RecognizerResults Recognize(AudioRecording audioRecording, Confi
/// <param name="outputDirectory">where results are to be put.</param>
/// <param name="segmentStartOffset">where one segment is located in the total recording.</param>
/// <returns>a list of events.</returns>
internal static RecognizerResults Gruntwork(AudioRecording audioRecording, Config configuration, DirectoryInfo outputDirectory, TimeSpan segmentStartOffset)
internal static RecognizerResults TerritorialCall(AudioRecording audioRecording, Config configuration, Config profile, DirectoryInfo outputDirectory, TimeSpan segmentStartOffset)
{
// get the common properties
string speciesName = configuration[AnalysisKeys.SpeciesName] ?? "Pteropus species";
string abbreviatedSpeciesName = configuration[AnalysisKeys.AbbreviatedSpeciesName] ?? "Pteropus";

// The following parameters worked well on a ten minute recording containing 14-16 calls.
// Note: if you lower the dB threshold, you need to increase maxDurationSeconds
int minHz = configuration.GetIntOrNull(AnalysisKeys.MinHz) ?? 800;
int maxHz = configuration.GetIntOrNull(AnalysisKeys.MaxHz) ?? 8000;
double minDurationSeconds = configuration.GetIntOrNull(AnalysisKeys.MinDuration) ?? 0.15;
double maxDurationSeconds = configuration.GetIntOrNull(AnalysisKeys.MaxDuration) ?? 0.5;
int minHz = profile.GetIntOrNull(AnalysisKeys.MinHz) ?? 800;
int maxHz = profile.GetIntOrNull(AnalysisKeys.MaxHz) ?? 8000;
double minDurationSeconds = profile.GetDoubleOrNull(AnalysisKeys.MinDuration) ?? 0.15;
double maxDurationSeconds = profile.GetDoubleOrNull(AnalysisKeys.MaxDuration) ?? 0.5;
double decibelThreshold = profile.GetDoubleOrNull(AnalysisKeys.DecibelThreshold) ?? 9.0;

var minTimeSpan = TimeSpan.FromSeconds(minDurationSeconds);
var maxTimeSpan = TimeSpan.FromSeconds(maxDurationSeconds);
double decibelThreshold = configuration.GetDoubleOrNull(AnalysisKeys.NoiseBgThreshold) ?? 9.0;

//######################
//2.Convert each segment to a spectrogram. Don't use samples in this recogniser.
Expand All @@ -131,7 +204,7 @@ internal static RecognizerResults Gruntwork(AudioRecording audioRecording, Confi
double intensityNormalisationMax = 3 * decibelThreshold;
var eventThreshold = decibelThreshold / intensityNormalisationMax;
var normalisedIntensityArray = DataTools.NormaliseInZeroOne(decibelArray, 0, intensityNormalisationMax);
var plot = new Plot(speciesName, normalisedIntensityArray, eventThreshold);
var plot = new Plot(speciesName + " Territory", normalisedIntensityArray, eventThreshold);
var plots = new List<Plot> { plot };

//iii: CONVERT decibel SCORES TO ACOUSTIC EVENTS
Expand All @@ -158,17 +231,6 @@ internal static RecognizerResults Gruntwork(AudioRecording audioRecording, Confi

acousticEvents = FilterEventsForSpectralProfile(acousticEvents, sonogram);

//Set following true if you want special debug spectrogram, i.e. with special plots
//In addition, standard spectrograms are produced when you set true in the config file, Towsey.PteropusSpecies.yml.
if (false)
{
//var image = sonogram.GetImageFullyAnnotated("Test");
var image = SpectrogramTools.GetSonogramPlusCharts(sonogram, acousticEvents, plots, null);
var opPath = outputDirectory.Combine(FilenameHelpers.AnalysisResultName(Path.GetFileNameWithoutExtension(audioRecording.BaseName), speciesName, "png", "DebugSpectrogram"));
string imageFilename = audioRecording.BaseName + ".png";
image.Save(Path.Combine(outputDirectory.FullName, imageFilename));
}

return new RecognizerResults()
{
Events = acousticEvents,
Expand All @@ -189,14 +251,19 @@ private static List<AcousticEvent> FilterEventsForSpectralProfile(List<AcousticE
{
double[,] spectrogramData = sonogram.Data;
int colCount = spectrogramData.GetLength(1);

// The following freq bins are used to demarcate freq bands for spectral tests below.
// The hertz values are hard coded but could be included in the config.yml file.
int maxBin = (int)Math.Round(8000 / sonogram.FBinWidth);
int fourkHzBin = (int)Math.Round(4000 / sonogram.FBinWidth);
int onekHzBin = (int)Math.Round(1000 / sonogram.FBinWidth);

var filteredEvents = new List<AcousticEvent>();
foreach (AcousticEvent ae in events)
{
int startFrame = ae.Oblong.RowTop;
int endFrame = ae.Oblong.RowBottom;

int maxBin = (int)Math.Round(8000 / sonogram.FBinWidth);

// get all the frames of the acoustic event
//var subMatrix = DataTools.Submatrix(spectrogramData, startFrame, 0, endFrame, colCount - 1);

Expand All @@ -212,11 +279,9 @@ private static List<AcousticEvent> FilterEventsForSpectralProfile(List<AcousticE
// Do TESTS to determine if event has spectrum matching a Flying fox.

// Test 1: Spectral maximum should be below 4 kHz.
int fourkHzBin = (int)Math.Round(4000 / sonogram.FBinWidth);
bool passTest1 = maxId < fourkHzBin;

// Test 2: There should be little energy in 0-1 kHz band.
int onekHzBin = (int)Math.Round(1000 / sonogram.FBinWidth);
var subband1Khz = DataTools.Subarray(normalisedSpectrum, 0, onekHzBin);
double bandArea1 = subband1Khz.Sum();
double energyRatio1 = bandArea1 / normalisedSpectrum.Sum();
Expand Down Expand Up @@ -256,5 +321,96 @@ private static List<AcousticEvent> FilterEventsForSpectralProfile(List<AcousticE

return filteredEvents;
}

/// <summary>
/// THis method does the work.
/// </summary>
/// <param name="audioRecording">the recording.</param>
/// <param name="configuration">the config file.</param>
/// <param name="outputDirectory">where results are to be put.</param>
/// <param name="segmentStartOffset">where one segment is located in the total recording.</param>
/// <returns>a list of events.</returns>
internal static RecognizerResults WingBeats(AudioRecording audioRecording, Config configuration, Config profile, DirectoryInfo outputDirectory, TimeSpan segmentStartOffset)
{
// get the common properties
string speciesName = configuration[AnalysisKeys.SpeciesName] ?? "Pteropus species";
string abbreviatedSpeciesName = configuration[AnalysisKeys.AbbreviatedSpeciesName] ?? "Pteropus";

// The following parameters worked well on a ten minute recording containing 14-16 calls.
// Note: if you lower the dB threshold, you need to increase maxDurationSeconds
int minHz = profile.GetIntOrNull(AnalysisKeys.MinHz) ?? 100;
int maxHz = profile.GetIntOrNull(AnalysisKeys.MaxHz) ?? 3000;
double minDurationSeconds = profile.GetDoubleOrNull(AnalysisKeys.MinDuration) ?? 1.0;
double maxDurationSeconds = profile.GetDoubleOrNull(AnalysisKeys.MaxDuration) ?? 10.0;
double dctDuration = profile.GetDoubleOrNull("DctDuration") ?? 1.0;
double dctThreshold = profile.GetDoubleOrNull("DctThreshold") ?? 0.5;
double minOscilFreq = profile.GetDoubleOrNull("MinOscilFreq") ?? 4.0;
double maxOscilFreq = profile.GetDoubleOrNull("MaxOscilFreq") ?? 6.0;
//var minTimeSpan = TimeSpan.FromSeconds(minDurationSeconds);
//var maxTimeSpan = TimeSpan.FromSeconds(maxDurationSeconds);
double eventThreshold = profile.GetDoubleOrNull("EventThreshold") ?? 0.3;

//######################
//2.Convert each segment to a spectrogram. Don't use samples in this recogniser.
//var samples = audioRecording.WavReader.Samples;

// make a spectrogram
var sonoConfig = new SonogramConfig
{
WindowSize = 512,
NoiseReductionType = NoiseReductionType.Standard,
NoiseReductionParameter = configuration.GetDoubleOrNull(AnalysisKeys.NoiseBgThreshold) ?? 0.0,
};
sonoConfig.WindowOverlap = 0.0;

// now construct the standard decibel spectrogram WITH noise removal, and look for LimConvex
// get frame parameters for the analysis
var sonogram = (BaseSonogram)new SpectrogramStandard(sonoConfig, audioRecording.WavReader);
var decibelArray = SNR.CalculateFreqBandAvIntensity(sonogram.Data, minHz, maxHz, sonogram.NyquistFrequency);

// Look for wing beats using oscillation detector
Oscillations2012.Execute(
(SpectrogramStandard)sonogram,
minHz,
maxHz,
dctDuration,
(int)Math.Floor(minOscilFreq),
(int)Math.Floor(maxOscilFreq),
dctThreshold,
eventThreshold,
minDurationSeconds,
maxDurationSeconds,
out var scores,
out var acousticEvents,
out var hits,
segmentStartOffset);

// prepare plots
double decibelThreshold = 12.0;
double intensityNormalisationMax = 3 * decibelThreshold;
var normThreshold = decibelThreshold / intensityNormalisationMax;
var normalisedIntensityArray = DataTools.NormaliseInZeroOne(decibelArray, 0, intensityNormalisationMax);
var plot1 = new Plot(speciesName + " Wingbeat band", normalisedIntensityArray, normThreshold);
var plot2 = new Plot(speciesName + " Wingbeat Osc Score", scores, eventThreshold);
var plots = new List<Plot> { plot1, plot2 };

// ######################################################################
acousticEvents.ForEach(ae =>
{
ae.SpeciesName = speciesName;
ae.SegmentDurationSeconds = audioRecording.Duration.TotalSeconds;
ae.SegmentStartSeconds = segmentStartOffset.TotalSeconds;
ae.Name = abbreviatedSpeciesName;
});

return new RecognizerResults()
{
Events = acousticEvents,
Hits = null,
ScoreTrack = null,
Plots = plots,
Sonogram = sonogram,
};
}
}
}

0 comments on commit 14a3214

Please sign in to comment.