Ircamdescriptor

see also Descripteurs

The ircamdescriptor set of tools implement a wide range of audio features to be used in content-based classification, clustering and processing applications. There are two parallel versions: one for Matlab, and one as a C++ library (the latter also available as a stand-alone command-line executable).

Most audio descriptors implemented are documented in Geoffroy Peeters' paper "A large set of audio features for sound description (similarity and classification) in the CUIDADO project" [1].

Contributors: Juan José Burred, Carmine Emanuele Cella, Geoffroy Peeters, Axel Röbel, Frédéric Cornu, Alessandro Saccoia, Patrice Tisserand

Official Internal Release

An shared internal release of the latest command line tool is accessible from the linux network only and is installed in the /u/formes/share directory. The required OS to use the program is a Red Hat 5.x kernel version. The program name is ircamdescriptor. A 64-bit Mac OSX version is also available for 10.6 or higher.

Latest Version	2.8.6
Last update	Mar 15, 2013
Documentation	Link to the documentation Configuration parameters Dependencies of the standard descriptors Dependencies of the energy descriptors
Access to the linux program	Analyse/Synthèse users To have all the paths correctly configured, you can manually call the /u/formes/share/etc/formes.cshrc script or put it in your .login script. Other teams The command line version of ircamdescriptor is installed in the shared folder of the analyse/synthèse subnetwork. Who wants to access the program should ask the sysadmins to setup an account for the linux network, unless there is one already (to try it is enough to open a ssh connection to a server with the usual ircam username/password and see if it accepts the login). An updated list of the machines of the equipe is available on this page: only machines labeled as "server" are free to login for external users, the other machines are personal workstations have less computing power and memory and should not be used remotely to avoid disturbing the user that is logged into a GUI session. To have all the paths correctly configured, you can manually call the /u/formes/share/etc/formes.cshrc script or put it in your .login script.
Mac OSX version	ircamdescriptor 2.8.6 for MacOS 10.6 (64bit)

Table of temporal modelings

	ircamdescriptor (Matlab)	ircamdescriptor (C++ new API)	SDIF Frame Type
ShortTermFeature			1DSC
(Loudness)WeightedMean	Variable length or File-based	Variable length or File-based	1WMN
(Loudness)WeightedStandardDeviation	Variable length or File-based	Variable length or File-based	1WSD
MedianFilter			1MED
TemporalFilterBankCoefficients			1TFB
Delta			1ODO
DeltaDelta			1OAO
Modulation			1MOD
ModulationAmplitude	Variable length or File-based	Variable length or File-based	1MDA
ModulationFrequency	Variable length or File-based	Variable length or File-based	1MDF
LogAttackTime	Variable length or File-based	Variable length or File-based	1LAT
TemporalIncrease	Variable length or File-based	Variable length or File-based	1TIN
TemporalDecrease	Variable length or File-based	Variable length or File-based	1TDE
TemporalCentroid	Variable length or File-based	Variable length or File-based	1TCN
EffectiveDuration	Variable length or File-based	Variable length or File-based	1EFD
Delta MedianFilter			1ODM
DeltaDelta MedianFilter			1OAM
WeightedMean MedianFilter			1MOM
WeightedStdDeviation MedianFilter			1SOM
TemporalFilterBankCoefficients MedianFilter			1TFM
WeightedMean Delta			1MDO
WeightedStdDeviation Delta			1SDO
TemporalFilterBankCoefficients Delta			1TDO
WeightedMean DeltaDelta			1MAO
WeightedStdDeviation DeltaDelta			1SAO
TemporalFilterBankCoefficients DeltaDelta			1TAO
WeightedMean Delta MedianFilter			1MDM
WeightedStdDeviation Delta MedianFilter			1SDM
TemporalFilterBankCoefficients Delta MedianFilter			1TDM
WeightedMean DeltaDelta MedianFilter			1MAM
WeightedStdDeviation DeltaDelta MedianFilter			1SAM
TemporalFilterBankCoefficients DeltaDelta MedianFilter			1TAM

MedianFilter (M), Delta (D), DeltaDelta (A), WeightedMean (M), WeightedStdDeviation(S), TemporalFilterBankCoefficients (T)

Table of descriptors

	dimensions	variations	Matlab field name	SDIF type	Matlab	C++
Temporal features
SignalAutoCorrelation	v[4](def.: 12)	1	i_xcorr_m	1ARR	Yes	Yes
SignalZeroCrossingRate	1	1	i_zcr_v	1ZCR	Yes	Yes
TimeFrame	1	1	i_timeframe_v	-	Yes [5]
Energy features
TotalEnergy	1	1	i_tot_v	1NRG	Yes	Yes
HarmonicEnergy	1	1	i_harmo_v	1HEN	Yes	Yes
NoiseEnergy	1	1	i_noise_v	1NEN	Yes	Yes
EnergyEnvelope	1	1	i_enenv_v	1EEV	Yes	Yes
Spectral features
SpectralCentroid	1	6	i_sc_v	1SCN	Yes	Yes
SpectralSpread	1	6	i_ss_v	1SSP	Yes	Yes
SpectralSkewness	1	6	i_skew_v	1SSK	Yes	Yes
SpectralKurtosis	1	6	i_kurto_v	1SKU	Yes	Yes
SpectralSlope	1	6	i_slope_v	1SSL	Yes	Yes
SpectralDecrease	1	1	i_decs_v	1SDE	Yes	Yes
SpectralRollOff	1	1	i_rolloff_v	1SRO	Yes	Yes
SpectralVariation	1	3	i_variation_v	1SVA	Yes	Yes
Harmonic features
FundamentalFrequency	1	1	i_f0_v	1FQ0	Yes	Yes
Noisiness	1	1	i_noisiness_v	1NSN	Yes	Yes
Inharmonicity	1	1	i_inharmo_v	1INH	Yes	Yes
HarmonicSpectralDeviation	1	3	i_devs_v	1HSD	Yes	Yes
HarmonicOddToEvenRatio	1	3	i_oeratio_v	1HOE	Yes	Yes
HarmonicTristimulus	3	3	i_tri_v	1HTR	Yes	Yes
HarmonicSpectralCentroid	1	6	i_sc_v	1HCN	Yes	Yes
HarmonicSpectralSpread	1	6	i_ss_v	1HSP	Yes	Yes
HarmonicSpectralSkewness	1	6	i_skew_v	1HSK	Yes	Yes
HarmonicSpectralKurtosis	1	6	i_kurto_v	1HKU	Yes	Yes
HarmonicSpectralSlope	1	6	i_slope_v	1HSL	Yes	Yes
HarmonicSpectralDecrease	1	1	i_decs_v	1HDE	Yes	Yes
HarmonicSpectralRollOff	1	1	i_rolloff_v	1HRO	Yes	Yes
HarmonicSpectralVariation	1	3	i_variation_v	1HVA	Yes	Yes
Perceptual features
PerceptualSpectralDeviation	1	3	i_devs_v	1PSD	Yes	Yes
PerceptualOddToEvenRatio	1	3	i_oeratio_v	1POE	Yes	Yes
PerceptualTristimulus	3	3	i_tri_v	1PTR	Yes	Yes
PerceptualSpectralCentroid	1	6	i_sc_v	1PCN	Yes	Yes
PerceptualSpectralSpread	1	6	i_ss_v	1PSP	Yes	Yes
PerceptualSpectralSkewness	1	6	i_skew_v	1PSK	Yes	Yes
PerceptualSpectralKurtosis	1	6	i_kurto_v	1PKU	Yes	Yes
PerceptualSpectralSlope	1	6	i_slope_v	1PSL	Yes	Yes
PerceptualSpectralDecrease	1	1	i_decs_v	1PDE	Yes	Yes
PerceptualSpectralRollOff	1	1	i_rolloff_v	1PRO	Yes	Yes
PerceptualSpectralVariation	1	3	i_variation_v	1PVA	Yes	Yes
Loudness	1	1	i_loud_v	1LDN	Yes	Yes
RelativeSpecificLoudness	v[4] (def.: 24)	1	i_specloud_m	1RSL	Yes	Yes
Sharpness	1	1	i_sharpness_v	1SHA	Yes	Yes
Spread	1	1	i_spread_v	1SPR	Yes	Yes
SpectralFlatness	4	1	i_sfm_m	1SFM	Yes	Yes
SpectralCrest	4	1	i_scm_m	1SCM	Yes	Yes
MFCC	v[4](def.: 13)	1	i_mfcc_m	1MFC	Yes	Yes
Chroma	(def.: 12)	1		1CHR

Noise shape features
NoiseSpectralCentroid	1	6	i_sc_v	1NCN	Yes
NoiseSpectralSpread	1	6	i_ss_v	1NSP	Yes
NoiseSpectralSkewness	1	6	i_skew_v	1NSK	Yes
NoiseSpectralKurtosis	1	6	i_kurto_v	1NKU	Yes
NoiseSpectralSlope	1	6	i_slope_v	1NSL	Yes
NoiseSpectralDecrease	1	1	i_decs_v	1NDE	Yes
NoiseSpectralRollOff	1	1	i_rolloff_v	1NRO	Yes
NoiseSpectralVariation	1	3	i_variation_v	1NVA	Yes
ERB-based features
BandFluctuationStrength	24	1	g_flustr_v	1BFL	Yes
BandRoughness	24	1	g_roughn_v	1BRG	Yes
FluctuationStrength	1	1	g_flustr	1FLS	Yes
Roughness	1	1	g_roughn	1RGH	Yes

[4] Variable (user-defined). Default value indicated in parentheses.

[5] Saved only in Matlab structure for reference. Should not be used as descriptor.

Old Releases Program specifications

	ircamdescriptor (Matlab)	IrcamDescriptor (module and template library) (C++ new API)
Latest version	0.44 (8/06/2009)	2.0 (03/09/2008)
Download	ircamdescriptor_matlab_v0.44.zip (Matlab p-code)	compiled library and standalone (ircamdescriptor) for OSX UB compiled library and standalone (ircamdescriptor) for Windows source code (SVN trac server)
Version History	changeLog_ircamdescriptor.txt	See source code
Documentation	ircamdescriptor_matlab_doc.pdf	API documentation
Input sound formats	WAV, AIFF	WAV, AIFF
Additional input required	.f0.sdif fundamental frequency file [1] .part.sdif sinusoidal analysis file [1]	None
Output formats	SDIF [8] MAT (Matlab structure) TXT (only for temporal models)	SDIF [8] RAW
Dependencies	loadsdif [1] writesdif [7] Note: sdif-matlab version 0.3.0.id needed. Download here (Mac OS, Linux 64)	None for standalones; IrcamSupport, libfft, libbpf, libf0, libpm2, libfip (only release mode), matmtl for libraries; Note: You can find a compiled version of the required libraries for Mac OSX and for Windows here
Send comments and bug reports to	burred@ircam.fr	cella@ircam.fr

[1] Only if harmonic or noise descriptors computed.

[7] Only if SDIF output selected.

[8] Extended SDIF types for Audio Descriptors: http://sdif.sourceforge.net/descriptor-types/

Matlab note: Older versions for Matlab can be found under /u/formes/share/packages/matlab/ircamdescriptor