jamie@1: /* libxtract feature extraction library
jamie@1: *
jamie@1: * Copyright (C) 2006 Jamie Bullock
jamie@1: *
jamie@1: * This program is free software; you can redistribute it and/or modify
jamie@1: * it under the terms of the GNU General Public License as published by
jamie@1: * the Free Software Foundation; either version 2 of the License, or
jamie@1: * (at your option) any later version.
jamie@1: *
jamie@1: * This program is distributed in the hope that it will be useful,
jamie@1: * but WITHOUT ANY WARRANTY; without even the implied warranty of
jamie@1: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
jamie@1: * GNU General Public License for more details.
jamie@1: *
jamie@1: * You should have received a copy of the GNU General Public License
jamie@1: * along with this program; if not, write to the Free Software
jamie@1: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
jamie@1: * USA.
jamie@1: */
jamie@1:
jamie@20: /** \mainpage
jamie@20: *
jamie@20: * LibXtract is a simple, portable, lightweight library of audio feature extraction functions. The purpose of the library is to provide a relatively exhaustive set of feature extraction primatives that are designed to be 'cascaded' to create a extraction hierarchies.
jamie@20: * For example, 'variance', 'average deviation', 'skewness' and 'kurtosis', all require the 'mean' of the input vector to be precomputed. However, rather than compute the 'mean' 'inside' each function, it is expected that the 'mean' will be passed in as an argument. This means that if the user wishes to use all of these features, the mean is calculated only once, and then passed to any functions that require it.
jamie@20: *
jamie@20: * This philosophy of 'cascading' features is followed throughout the library, for example with features that operate on the magnitude spectrum of a signal vector (e.g. 'irregularity'), the magnitude spectrum is not calculated 'inside' the respective function, instead, a pointer to the first element in an array containing the magnitude spectrum is passed in as an argument.
jamie@20: *
jamie@105: * Hopefully this not only makes the library more efficient when computing large numbers of features, but also makes it more flexible because extraction functions can be combined arbitrarily (one can take the irregularility of the Mel Frequency Cepstral Coefficients for example).
jamie@20: *
jamie@105: * All feature extraction functions follow the same prototype:
jamie@105: *
jamie@105: int xtract_function_name(const float *data, const int N, const void *argv, float *result){
jamie@105: *
jamie@105: * \param const float *data points to an array of floats representing the input data
jamie@105: * \param const int N represents the number of elementes from *data to be considered in the calculation
jamie@105: * \param const void *argv represents an arbitrary list of arguments. Used to pass in values required by the feature calculation
jamie@105: * \param float *result points to an array of floats, or a single float represnting the result of the calculation
jamie@105: *
jamie@105: *
jamie@105: * It is up to the calling function to allocate enough memory for the *data, *argv, and *result, and to free it when required. Some feature extraction functions may also require an _init() function to be called in order to perform some initialisation. The struct xtract_function_descriptor_t is used to give an indication of recommended default values, and argc for the *argv array.
jamie@83: *
jamie@83: * LibXtract can be downloaded from http://www.sf.net/projects/libxtract
jamie@83: *
jamie@20: */
jamie@20:
jamie@1: #ifndef XTRACT_H
jamie@1: #define XTRACT_H
jamie@1:
jamie@1: #ifdef __cplusplus
jamie@1: extern "C" {
jamie@1: #endif
jamie@1:
jamie@21: /**
jamie@21: * \file libxtract.h
jamie@21: * \brief main header file and API definition
jamie@1: */
jamie@1:
jamie@1: #include "xtract_scalar.h"
jamie@1: #include "xtract_vector.h"
jamie@1: #include "xtract_delta.h"
jamie@1: #include "xtract_types.h"
jamie@1: #include "xtract_macros.h"
jamie@1:
jamie@20: /** \defgroup libxtract API
jamie@20: *
jamie@20: * Defines a very simple API that provides access to the functions in the library
jamie@20: * @{
jamie@20: */
jamie@20:
jamie@106: #define XTRACT_FEATURES 57
jamie@30:
jamie@1: /** \brief Enumeration of features, elements are used as indixes to an array of pointers to feature extracton functions */
jamie@56: enum xtract_features_ {
jamie@56: XTRACT_MEAN,
jamie@56: XTRACT_VARIANCE,
jamie@56: XTRACT_STANDARD_DEVIATION,
jamie@56: XTRACT_AVERAGE_DEVIATION,
jamie@56: XTRACT_SKEWNESS,
jamie@56: XTRACT_KURTOSIS,
jamie@56: XTRACT_SPECTRAL_MEAN,
jamie@56: XTRACT_SPECTRAL_VARIANCE,
jamie@56: XTRACT_SPECTRAL_STANDARD_DEVIATION,
jamie@56: XTRACT_SPECTRAL_AVERAGE_DEVIATION,
jamie@56: XTRACT_SPECTRAL_SKEWNESS,
jamie@56: XTRACT_SPECTRAL_KURTOSIS,
jamie@56: XTRACT_SPECTRAL_CENTROID,
jamie@56: XTRACT_IRREGULARITY_K,
jamie@56: XTRACT_IRREGULARITY_J,
jamie@56: XTRACT_TRISTIMULUS_1,
jamie@56: XTRACT_TRISTIMULUS_2,
jamie@56: XTRACT_TRISTIMULUS_3,
jamie@56: XTRACT_SMOOTHNESS,
jamie@56: XTRACT_SPREAD,
jamie@56: XTRACT_ZCR,
jamie@56: XTRACT_ROLLOFF,
jamie@56: XTRACT_LOUDNESS,
jamie@56: XTRACT_FLATNESS,
jamie@56: XTRACT_TONALITY,
jamie@56: XTRACT_CREST,
jamie@56: XTRACT_NOISINESS,
jamie@56: XTRACT_RMS_AMPLITUDE,
jamie@56: XTRACT_SPECTRAL_INHARMONICITY,
jamie@56: XTRACT_POWER,
jamie@56: XTRACT_ODD_EVEN_RATIO,
jamie@56: XTRACT_SHARPNESS,
jamie@56: XTRACT_SPECTRAL_SLOPE,
jamie@56: XTRACT_LOWEST_VALUE,
jamie@56: XTRACT_HIGHEST_VALUE,
jamie@56: XTRACT_SUM,
jamie@59: XTRACT_NONZERO_COUNT,
jamie@56: XTRACT_HPS,
jamie@56: XTRACT_F0,
jamie@56: XTRACT_FAILSAFE_F0,
jamie@106: XTRACT_LNORM,
jamie@56: XTRACT_FLUX,
jamie@56: XTRACT_ATTACK_TIME,
jamie@56: XTRACT_DECAY_TIME,
jamie@106: XTRACT_DIFFERENCE_VECTOR,
jamie@56: XTRACT_AUTOCORRELATION,
jamie@56: XTRACT_AMDF,
jamie@56: XTRACT_ASDF,
jamie@56: XTRACT_BARK_COEFFICIENTS,
jamie@56: XTRACT_PEAK_SPECTRUM,
jamie@56: XTRACT_SPECTRUM,
jamie@56: XTRACT_AUTOCORRELATION_FFT,
jamie@56: XTRACT_MFCC,
jamie@56: XTRACT_DCT,
jamie@104: XTRACT_HARMONIC_SPECTRUM,
jamie@104: XTRACT_LPC,
jamie@104: XTRACT_LPCC
jamie@1: };
jamie@1:
jamie@55: /** \brief Enumeration of feature initialisation functions */
jamie@56: enum xtract_feature_init_ {
jamie@56: XTRACT_INIT_MFCC = 100,
jamie@56: XTRACT_INIT_BARK
jamie@55: };
jamie@55:
jamie@1: /** \brief Enumeration of feature types */
jamie@56: enum xtract_feature_types_ {
jamie@56: XTRACT_SCALAR,
jamie@56: XTRACT_VECTOR,
jamie@56: XTRACT_DELTA
jamie@1: };
jamie@1:
jamie@1: /** \brief Enumeration of mfcc types */
jamie@56: enum xtract_mfcc_types_ {
jamie@56: XTRACT_EQUAL_GAIN,
jamie@56: XTRACT_EQUAL_AREA
jamie@1: };
jamie@1:
jamie@106: enum xtract_lnorm_filter_types_ {
jamie@106: XTRACT_POSITIVE_SLOPE,
jamie@106: XTRACT_NEGATIVE_SLOPE
jamie@106: };
jamie@106:
jamie@1: /** \brief Enumeration of return codes */
jamie@56: enum xtract_return_codes_ {
jamie@56: XTRACT_SUCCESS,
jamie@56: XTRACT_MALLOC_FAILED,
jamie@56: XTRACT_BAD_ARGV,
jamie@56: XTRACT_BAD_VECTOR_SIZE,
jamie@56: XTRACT_NO_RESULT,
jamie@56: XTRACT_FEATURE_NOT_IMPLEMENTED
jamie@1: };
jamie@1:
jamie@54: /** \brief Enumeration of spectrum types */
jamie@56: enum xtract_spectrum_ {
jamie@56: XTRACT_MAGNITUDE_SPECTRUM,
jamie@56: XTRACT_LOG_MAGNITUDE_SPECTRUM,
jamie@56: XTRACT_POWER_SPECTRUM,
jamie@56: XTRACT_LOG_POWER_SPECTRUM
jamie@54: };
jamie@54:
jamie@50: /** \brief Enumeration of data types*/
jamie@50: typedef enum type_ {
jamie@56: XTRACT_FLOAT,
jamie@56: XTRACT_FLOATARRAY,
jamie@56: XTRACT_INT,
jamie@56: XTRACT_MEL_FILTER
jamie@56: } xtract_type_t;
jamie@50:
jamie@50: /** \brief Enumeration of units*/
jamie@50: typedef enum unit_ {
jamie@55: /* NONE, ANY */
jamie@56: XTRACT_HERTZ = 2,
jamie@56: XTRACT_ANY_AMPLITUDE_HERTZ,
jamie@56: XTRACT_DBFS,
jamie@56: XTRACT_DBFS_HERTZ,
jamie@56: XTRACT_PERCENT,
jamie@56: XTRACT_SONE
jamie@56: } xtract_unit_t;
jamie@50:
jamie@50: /** \brief Boolean */
jamie@50: typedef enum {
jamie@56: XTRACT_FALSE,
jamie@56: XTRACT_TRUE
jamie@56: } xtract_bool_t;
jamie@50:
jamie@50: /** \brief Enumeration of vector format types*/
jamie@56: typedef enum xtract_vector_ {
jamie@54: /* N/2 magnitude/log-magnitude/power/log-power coeffs and N/2 frequencies */
jamie@56: XTRACT_SPECTRAL,
jamie@54: /* N spectral amplitudes */
jamie@56: XTRACT_SPECTRAL_MAGNITUDES,
jamie@54: /* N/2 magnitude/log-magnitude/power/log-power peak coeffs and N/2
jamie@54: * frequencies */
jamie@56: XTRACT_SPECTRAL_PEAKS,
jamie@54: /* N spectral peak amplitudes */
jamie@59: XTRACT_SPECTRAL_PEAKS_MAGNITUDES,
jamie@59: /* N spectral peak frequencies */
jamie@59: XTRACT_SPECTRAL_PEAKS_FREQUENCIES,
jamie@54: /* N/2 magnitude/log-magnitude/power/log-power harmonic peak coeffs and N/2
jamie@54: * frequencies */
jamie@56: XTRACT_SPECTRAL_HARMONICS,
jamie@54: /* N spectral harmonic amplitudes */
jamie@56: XTRACT_SPECTRAL_HARMONICS_MAGNITUDES,
jamie@54: /* N spectral harmonic frequencies */
jamie@56: XTRACT_SPECTRAL_HARMONICS_FREQUENCIES,
jamie@104: XTRACT_AUTOCORRELATION_COEFFS,
jamie@56: XTRACT_ARBITRARY_SERIES,
jamie@56: XTRACT_AUDIO_SAMPLES,
jamie@56: XTRACT_MEL_COEFFS,
jamie@104: XTRACT_LPC_COEFFS,
jamie@104: XTRACT_LPCC_COEFFS,
jamie@56: XTRACT_BARK_COEFFS,
jamie@56: XTRACT_NO_DATA
jamie@56: } xtract_vector_t;
jamie@50:
jamie@50: /** \brief Data structure containing useful information about functions provided by LibXtract. */
jamie@56: typedef struct _xtract_function_descriptor {
jamie@50:
jamie@50: struct {
jamie@56: char name[XTRACT_MAX_NAME_LENGTH];
jamie@56: char p_name[XTRACT_MAX_NAME_LENGTH]; /* pretty name */
jamie@56: char desc[XTRACT_MAX_DESC_LENGTH];
jamie@56: char p_desc[XTRACT_MAX_DESC_LENGTH]; /* pretty description */
jamie@56: char author[XTRACT_MAX_AUTHOR_LENGTH];
jamie@50: int year;
jamie@50: } algo;
jamie@50:
jamie@50: struct {
jamie@56: xtract_vector_t format;
jamie@56: xtract_unit_t unit;
jamie@50: } data;
jamie@50:
jamie@51: int argc;
jamie@50:
jamie@50: struct {
jamie@56: xtract_type_t type; /* type of the array/value pointed to by argv */
jamie@56: float min[XTRACT_MAXARGS];
jamie@56: float max[XTRACT_MAXARGS];
jamie@56: float def[XTRACT_MAXARGS]; /* defaults */
jamie@56: xtract_unit_t unit[XTRACT_MAXARGS];
jamie@56: int donor[XTRACT_MAXARGS]; /* suggested donor functions for argv */
jamie@50: } argv;
jamie@50:
jamie@56: xtract_bool_t is_scalar;
jamie@50:
jamie@55: /* The result.<> entries in descritors.c need to be checked */
jamie@50: union {
jamie@50:
jamie@50: struct {
jamie@50: float min;
jamie@50: float max;
jamie@56: xtract_unit_t unit;
jamie@50: } scalar;
jamie@50:
jamie@50: struct {
jamie@56: xtract_vector_t format;
jamie@56: xtract_unit_t unit;
jamie@50: } vector;
jamie@50:
jamie@50: } result;
jamie@50:
jamie@56: } xtract_function_descriptor_t;
jamie@50:
jamie@1: /**
jamie@1: *
jamie@2: * \brief An array of pointers to functions that perform the extraction
jamie@1: *
jamie@2: * \param *data: a pointer to the start of the input data (usually the first element in an array)
jamie@1: *
jamie@2: * \param N: the number of elements to be processed
jamie@1: *
jamie@98: * \param *argv: an abitrary number of additional arguments, used to pass additional parameters to the function being called. All arguments are compulsary!
jamie@1: *
jamie@2: * \param *result: a pointer to the first element in the result
jamie@1: *
jamie@1: * Each function will iterate over N array elements, the first of which is
jamie@2: * pointed to by *data. It is up to the calling function to ensure that the array is in the format expected by the function being called.
jamie@1: *
jamie@1: * For scalar and delta features, *result will point to a single value.
jamie@1: *
jamie@1: * For vector features it will point to the first element in an array.
jamie@1: *
jamie@1: * Memory for this array must be allocated and freed by the calling
jamie@1: * function.
jamie@1: *
jamie@21: * All functions return an integer error code as descibed in the enumeration
jamie@21: * return_codes_
jamie@2: *
jamie@21: * The preprocessor macro: XTRACT must be defined before this can be used
jamie@9: *
jamie@2: * example:
jamie@21: * \verbatim
jamie@21: #include
jamie@21: #define XTRACT
jamie@21: #include "libxtract.h"
jamie@21:
jamie@21: main () {
jamie@21: float values[] = {1.0, 2.0, 3.0, 4.0, 5.0};
jamie@21: int N = 5;
jamie@21: float mean;
jamie@21:
jamie@21: xtract[MEAN]((void *)values, N, NULL, &mean);
jamie@21:
jamie@21: printf("Mean = %.2f\n", mean);
jamie@21: }
jamie@21: \endverbatim
jamie@9: * The calling function may additionally make some tests against the value returned by xtract
jamie@9: *
jamie@2: */
jamie@56: #ifdef XTRACT_H
jamie@43: extern int(*xtract[XTRACT_FEATURES])(const float *data, const int N, const void *argv, float *result);
jamie@1:
jamie@28: #endif
jamie@26:
jamie@2: /** \brief A structure to store a set of n_filters Mel filters */
jamie@1: typedef struct xtract_mel_filter_ {
jamie@1: int n_filters;
jamie@1: float **filters;
jamie@1: } xtract_mel_filter;
jamie@1:
jamie@2: /** \brief A function to initialise a mel filter bank
jamie@2: *
jamie@2: * It is up to the caller to pass in a pointer to memory allocated for freq_bands arrays of length N. This function populates these arrays with magnitude coefficients representing the mel filterbank on a linear scale
jamie@2: */
jamie@56: int xtract_init_mfcc(int N, float nyquist, int style, float freq_min, float freq_max, int freq_bands, float **fft_tables);
jamie@1:
jamie@2: /** \brief A function to initialise bark filter bounds
jamie@2: *
jamie@2: * A pointer to an array of BARK_BANDS ints most be passed in, and is populated with BARK_BANDS fft bin numbers representing the limits of each band
jamie@59: *
jamie@59: * \param N: the audio block size
jamie@59: * \param sr: The sample audio sample rate
jamie@59: * \param *band_limits: a pointer to an array of BARK_BANDS ints
jamie@2: */
jamie@59: int xtract_init_bark(int N, float sr, int *band_limits);
jamie@1:
jamie@98: /** \brief An initialisation function for functions using FFT
jamie@98: *
jamie@98: * This function initialises global data structures used by functions requiring FFT functionality. It can be called multiple times with different feature names. Calling it more than once with the same feature name is not a valid operation and will result in a memory leak.
jamie@98: *
jamie@98: * \param N: the size of the FFT
jamie@98: * \param feature_name: the name of the feature the FFT is being used for,
jamie@98: * e.g. XTRACT_DCT
jamie@98: *
jamie@98: */
jamie@98: int xtract_init_fft(int N, int feature_name);
jamie@98:
jamie@50: /* \brief A function to build an array of function descriptors */
jamie@50: void *xtract_make_descriptors();
jamie@1:
jamie@50: /* \brief A function to free an array of function descriptors */
jamie@50: int xtract_free_descriptors(void *fd);
jamie@1: /* Free functions */
jamie@1:
jamie@20: /** @} */
jamie@20:
jamie@1: #ifdef __cplusplus
jamie@1: }
jamie@1: #endif
jamie@1:
jamie@1: #endif