view src/vector.c @ 214:f28f66faa016

Add "stateful" feature type with initial feature "last n"

Stateful feature extraction functions are functions that require state to be maintained between successive calls.  This is necessary, for example when an accumulation of values is required, or changes need to be measured over time.

The initial xtract_last_n() function accumulates the last N (single) values from *data and writes them to *result

/*
 * Copyright (C) 2012 Jamie Bullock
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to
 * deal in the Software without restriction, including without limitation the
 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
 * sell copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 */

/* xtract_vector.c: defines functions that extract a feature as a single value from an input vector */

#include <math.h>
#include <string.h>
#include <stdlib.h>
#include <assert.h>

#include "fft.h"

#include "../xtract/libxtract.h"
#include "xtract_macros_private.h"
#include "xtract_globals_private.h"

#ifndef M_PI
#define M_PI 3.14159265358979323846264338327
#endif

int xtract_spectrum(const double *data, const int N, const void *argv, double *result)
{

    int vector     = 0;
    int withDC     = 0;
    int normalise  = 0;
    double q        = 0.0;
    double temp     = 0.0;
    double max      = 0.0;
    double NxN      = XTRACT_SQ(N);
    double *marker  = NULL;
    double real = 0.0;
    double imag = 0.0;
    unsigned int n = 0;
    unsigned int m = 0;
    unsigned int M = N >> 1;
#ifdef USE_OOURA
    double *fft = NULL;
#else 
    DSPDoubleSplitComplex *fft = NULL;
#endif

    q = *(double *)argv;
    vector = (int)*((double *)argv+1);
    withDC = (int)*((double *)argv+2);
    normalise = (int)*((double *)argv+3);

    XTRACT_CHECK_q;
#ifdef USE_OOURA
    if(!ooura_data_spectrum.initialised)
#else
    if(!vdsp_data_spectrum.initialised)
#endif
    {
        fprintf(stderr,
                "libxtract: error: xtract_spectrum() failed, "
                "fft data unitialised.\n");
        return XTRACT_NO_RESULT;
    }

#ifdef USE_OOURA
    /* ooura is in-place
     * the output format is
     * a[0] - DC, a[1] - nyquist, a[2...N-1] - remaining bins
     */
    fft = (double*)malloc(N * sizeof(double));
    assert(fft != NULL);
    memcpy(fft, data, N * sizeof(double));

    rdft(N, 1, fft, ooura_data_spectrum.ooura_ip, 
            ooura_data_spectrum.ooura_w);
#else
    fft = &vdsp_data_spectrum.fft;
    vDSP_ctozD((DSPDoubleComplex *)data, 2, fft, 1, N >> 1);
    vDSP_fft_zripD(vdsp_data_spectrum.setup, fft, 1, 
            vdsp_data_spectrum.log2N, FFT_FORWARD);
#endif

    switch(vector)
    {

        case XTRACT_LOG_MAGNITUDE_SPECTRUM:
        for(n = 0, m = 0; m < M; ++n, ++m)
        {
            marker = &result[m];

            if(n==0 && !withDC) /* discard DC and keep Nyquist */
            {
                ++n;
#ifdef USE_OOURA
                marker = &result[M-1];
#endif
            }
#ifdef USE_OOURA
            if(n==1 && withDC) /* discard Nyquist */
            {
                ++n;
            }
            if(n == M)
            {
                XTRACT_SET_FREQUENCY;
                break;
            }

            real = fft[n*2];
            imag = fft[n*2+1];
#else
            real = fft->realp[n];
            imag = fft->realp[n];
#endif

            temp = XTRACT_SQ(real) + XTRACT_SQ(imag);
            if (temp > XTRACT_LOG_LIMIT)
            {
                temp = log(sqrt(temp) / (double)N);
            }
            else
            {
                temp = XTRACT_LOG_LIMIT_DB;
            }
            result[m] =
                /* Scaling */
                (temp + XTRACT_DB_SCALE_OFFSET) /
                XTRACT_DB_SCALE_OFFSET;

            XTRACT_SET_FREQUENCY;
            XTRACT_GET_MAX;
        }
        break;

    case XTRACT_POWER_SPECTRUM:
        for(n = 0, m = 0; m < M; ++n, ++m)
        {

            marker = &result[m];

            if(n==0 && !withDC) /* discard DC and keep Nyquist */
            {
                ++n;
#ifdef USE_OOURA
                marker = &result[M-1];
#endif
            }
#ifdef USE_OOURA
            if(n==1 && withDC) /* discard Nyquist */
            {
                ++n;
            }
            if(n == M)
            {
                XTRACT_SET_FREQUENCY;
                break;
            }

            real = fft[n*2];
            imag = fft[n*2+1];
#else
            real = fft->realp[n];
            imag = fft->realp[n];
#endif

            result[m] = (XTRACT_SQ(real) + XTRACT_SQ(imag)) / NxN;
            XTRACT_SET_FREQUENCY;
            XTRACT_GET_MAX;
        }
        break;

    case XTRACT_LOG_POWER_SPECTRUM:
        for(n = 0, m = 0; m < M; ++n, ++m)
        {
            marker = &result[m];

            if(n==0 && !withDC) /* discard DC and keep Nyquist */
            {
                ++n;
#ifdef USE_OOURA
                marker = &result[M-1];
#endif
            }
#ifdef USE_OOURA
            if(n==1 && withDC) /* discard Nyquist */
            {
                ++n;
            }
            if(n == M)
            {
                XTRACT_SET_FREQUENCY;
                break;
            }

            real = fft[n*2];
            imag = fft[n*2+1];
#else
            real = fft->realp[n];
            imag = fft->realp[n];
#endif

            if ((temp = XTRACT_SQ(real) + XTRACT_SQ(imag)) >
                    XTRACT_LOG_LIMIT)
                temp = log(temp / NxN);
            else
                temp = XTRACT_LOG_LIMIT_DB;

            result[m] = (temp + XTRACT_DB_SCALE_OFFSET) /
                        XTRACT_DB_SCALE_OFFSET;
            XTRACT_SET_FREQUENCY;
            XTRACT_GET_MAX;
        }
        break;

    default:
        /* MAGNITUDE_SPECTRUM */
        for(n = 0, m = 0; m < M; ++n, ++m)
        {
            marker = &result[m];

            if(n==0 && !withDC) /* discard DC and keep Nyquist */
            {
                ++n;
#ifdef USE_OOURA
                marker = &result[M-1];
#endif
            }
#ifdef USE_OOURA
            if(n==1 && withDC) /* discard Nyquist */
            {
                ++n;
            }
            if(n == M)
            {
                XTRACT_SET_FREQUENCY;
                break;
            }

            real = fft[n*2];
            imag = fft[n*2+1];
#else
            real = fft->realp[n];
            imag = fft->realp[n];
#endif
            *marker = sqrt(XTRACT_SQ(real) + XTRACT_SQ(imag)) / (double)N;

            XTRACT_SET_FREQUENCY;
            XTRACT_GET_MAX;
        }
        break;
    }

    if(normalise)
    {
        for(n = 0; n < M; n++)
            result[n] /= max;
    }

#ifdef USE_OOURA
    free(fft);
#endif

    return XTRACT_SUCCESS;
}

int xtract_autocorrelation_fft(const double *data, const int N, const void *argv, double *result)
{

    double *rfft = NULL;
    int n        = 0;
    int M        = 0;
#ifndef USE_OOURA
    DSPDoubleSplitComplex *fft = NULL;
    double M_double = 0.0;
#endif

    M = N << 1;

    /* Zero pad the input vector */
    rfft = (double *)calloc(M, sizeof(double));
    memcpy(rfft, data, N * sizeof(double));

#ifdef USE_OOURA
    rdft(M, 1, rfft, ooura_data_autocorrelation_fft.ooura_ip, 
            ooura_data_autocorrelation_fft.ooura_w);

    for(n = 2; n < M; ++n)
    {
        rfft[n*2] = XTRACT_SQ(rfft[n*2]) + XTRACT_SQ(rfft[n*2+1]);
        rfft[n*2+1] = 0.0;
    }

    rfft[0] = XTRACT_SQ(rfft[0]);
    rfft[1] = XTRACT_SQ(rfft[1]);

    rdft(M, -1, rfft, ooura_data_autocorrelation_fft.ooura_ip,
            ooura_data_autocorrelation_fft.ooura_w);

#else
    /* vDSP has its own autocorrelation function, but it doesn't fit the 
     * LibXtract model, e.g. we can't guarantee it's going to use
     * an FFT for all values of N */
    fft = &vdsp_data_autocorrelation_fft.fft;
    vDSP_ctozD((DSPDoubleComplex *)data, 2, fft, 1, N);
    vDSP_fft_zripD(vdsp_data_autocorrelation_fft.setup, fft, 1, 
            vdsp_data_autocorrelation_fft.log2N, FFT_FORWARD);

    for(n = 0; n < N; ++n)
    {
        fft->realp[n] = XTRACT_SQ(fft->realp[n]) + XTRACT_SQ(fft->imagp[n]);
        fft->imagp[n] = 0.0;
    }

    vDSP_fft_zripD(vdsp_data_autocorrelation_fft.setup, fft, 1, 
            vdsp_data_autocorrelation_fft.log2N, FFT_INVERSE);
#endif

    /* Normalisation factor */
    M = M * N;

#ifdef USE_OOURA
    for(n = 0; n < N; n++)
        result[n] = rfft[n] / (double)M;
    free(rfft);
#else
    M_double = (double)M;
    vDSP_ztocD(fft, 1, (DOUBLE_COMPLEX *)result, 2, N);
    vDSP_vsdivD(result, 1, &M_double, result, 1, N);
#endif

    return XTRACT_SUCCESS;
}

int xtract_mfcc(const double *data, const int N, const void *argv, double *result)
{

    xtract_mel_filter *f;
    int n, filter;

    f = (xtract_mel_filter *)argv;

    for(filter = 0; filter < f->n_filters; filter++)
    {
        result[filter] = 0.0;
        for(n = 0; n < N; n++)
        {
            result[filter] += data[n] * f->filters[filter][n];
        }
        result[filter] = log(result[filter] < XTRACT_LOG_LIMIT ? XTRACT_LOG_LIMIT : result[filter]);
    }

    xtract_dct(result, f->n_filters, NULL, result);

    return XTRACT_SUCCESS;
}

int xtract_dct(const double *data, const int N, const void *argv, double *result)
{

    int n;
    int m;
    double *temp = (double*)calloc(N, sizeof(double));

    for (n = 0; n < N; ++n)
    {
        for(m = 1; m <= N; ++m) {
            temp[n] += data[m - 1] * cos(M_PI * (n / (double)N) * (m - 0.5));
        }
    }

    memcpy(result, temp, N * sizeof(double));
    free(temp);

    return XTRACT_SUCCESS;
}

int xtract_autocorrelation(const double *data, const int N, const void *argv, double *result)
{

    /* Naive time domain implementation  */

    int n = N, i;

    double corr;

    while(n--)
    {
        corr = 0;
        for(i = 0; i < N - n; i++)
        {
            corr += data[i] * data[i + n];
        }
        result[n] = corr / N;
    }

    return XTRACT_SUCCESS;
}

int xtract_amdf(const double *data, const int N, const void *argv, double *result)
{

    int n = N, i;

    double md, temp;

    while(n--)
    {
        md = 0.0;
        for(i = 0; i < N - n; i++)
        {
            temp = data[i] - data[i + n];
            temp = (temp < 0 ? -temp : temp);
            md += temp;
        }
        result[n] = md / (double)N;
    }

    return XTRACT_SUCCESS;
}

int xtract_asdf(const double *data, const int N, const void *argv, double *result)
{

    int n = N, i;

    double sd;

    while(n--)
    {
        sd = 0.0;
        for(i = 0; i < N - n; i++)
        {
            /*sd = 1;*/
            sd += XTRACT_SQ(data[i] - data[i + n]);
        }
        result[n] = sd / (double)N;
    }

    return XTRACT_SUCCESS;
}

int xtract_bark_coefficients(const double *data, const int N, const void *argv, double *result)
{

    int *limits, band, n;

    limits = (int *)argv;

    for(band = 0; band < XTRACT_BARK_BANDS - 1; band++)
    {
        result[band] = 0.0;
        for(n = limits[band]; n < limits[band + 1]; n++)
            result[band] += data[n];
    }

    return XTRACT_SUCCESS;
}

int xtract_peak_spectrum(const double *data, const int N, const void *argv, double *result)
{

    double threshold, max, y, y2, y3, p, q, *input = NULL;
    size_t bytes;
    int n = N, rv = XTRACT_SUCCESS;

    threshold = max = y = y2 = y3 = p = q = 0.0;

    if(argv != NULL)
    {
        q = ((double *)argv)[0];
        threshold = ((double *)argv)[1];
    }
    else
        rv = XTRACT_BAD_ARGV;

    if(threshold < 0 || threshold > 100)
    {
        threshold = 0;
        rv = XTRACT_BAD_ARGV;
    }

    XTRACT_CHECK_q;

    input = (double *)calloc(N,  sizeof(double));

    bytes = N * sizeof(double);

    if(input != NULL)
        input = (double*)memcpy(input, data, bytes);
    else
        return XTRACT_MALLOC_FAILED;

    while(n--)
        max = XTRACT_MAX(max, input[n]);

    threshold *= .01 * max;

    result[0] = 0;
    result[N] = 0;

    for(n = 1; n < N; n++)
    {
        if(input[n] >= threshold)
        {
            if(input[n] > input[n - 1] && n + 1 < N && input[n] > input[n + 1])
            {
                result[N + n] = q * (n + 1 + (p = .5 * ((y = input[n-1]) -
                                                    (y3 = input[n+1])) / (input[n - 1] - 2 *
                                                            (y2 = input[n]) + input[n + 1])));
                result[n] = y2 - .25 * (y - y3) * p;
            }
            else
            {
                result[n] = 0;
                result[N + n] = 0;
            }
        }
        else
        {
            result[n] = 0;
            result[N + n] = 0;
        }
    }

    free(input);
    return (rv ? rv : XTRACT_SUCCESS);
}

int xtract_harmonic_spectrum(const double *data, const int N, const void *argv, double *result)
{

    int n = (N >> 1), M = n;

    const double *freqs, *amps;
    double f0, threshold, ratio, nearest, distance;

    amps = data;
    freqs = data + n;
    f0 = *((double *)argv);
    threshold = *((double *)argv+1);

    ratio = nearest = distance = 0.0;

    while(n--)
    {
        if(freqs[n])
        {
            ratio = freqs[n] / f0;
			nearest = floor( 0.5f + ratio);				// replace -> nearest = round(ratio);
			distance = fabs(nearest - ratio);
            if(distance > threshold)
                result[n] = result[M + n] = 0.0;
            else
            {
                result[n] = amps[n];
                result[M + n] = freqs[n];
            }
        }
        else
            result[n] = result[M + n] = 0.0;
    }
    return XTRACT_SUCCESS;
}

int xtract_lpc(const double *data, const int N, const void *argv, double *result)
{

    int i, j, M, L;
    double r = 0.0,
          error = 0.0;

    double *ref = NULL,
           *lpc = NULL ;

    error = data[0];
    L = N - 1; /* The number of LPC coefficients */
    M = L * 2; /* The length of *result */
    ref = result;
    lpc = result+L;

    if(error == 0.0)
    {
        memset(result, 0, M * sizeof(double));
        return XTRACT_NO_RESULT;
    }

    memset(result, 0, M * sizeof(double));

    for (i = 0; i < L; i++)
    {

        /* Sum up this iteration's reflection coefficient. */
        r = -data[i + 1];
        for (j = 0; j < i; j++)
            r -= lpc[j] * data[i - j];
        ref[i] = r /= error;

        /* Update LPC coefficients and total error. */
        lpc[i] = r;
        for (j = 0; j < i / 2; j++)
        {
            double tmp      = lpc[j];
            lpc[j]          = r * lpc[i - 1 - j];
            lpc[i - 1 - j] += r * tmp;
        }
        if (i % 2) lpc[j] += lpc[j] * r;

        error *= 1 - r * r;
    }

    return XTRACT_SUCCESS;
}

int xtract_lpcc(const double *data, const int N, const void *argv, double *result)
{

    /* Given N lpc coefficients extract an LPC cepstrum of size argv[0] */
    /* Based on an an algorithm by rabiner and Juang */

    int n, k;
    double sum;
    int order = N - 1; /* Eventually change this to Q = 3/2 p as suggested in Rabiner */
    int cep_length;

    if(argv == NULL)
        cep_length = N - 1; /* FIX: if we're going to have default values, they should come from the descriptor */
    else
        cep_length = *(int *)argv;
    //cep_length = (int)((double *)argv)[0];

    memset(result, 0, cep_length * sizeof(double));

    for (n = 1; n <= order && n <= cep_length; n++)
    {
        sum = 0.0;
        for (k = 1; k < n; k++)
            sum += k * result[k-1] * data[n - k];
        result[n-1] = data[n] + sum / n;
    }

    /* be wary of these interpolated values */
    for(n = order + 1; n <= cep_length; n++)
    {
        sum = 0.0;
        for (k = n - (order - 1); k < n; k++)
            sum += k * result[k-1] * data[n - k];
        result[n-1] = sum / n;
    }

    return XTRACT_SUCCESS;

}
//int xtract_lpcc_s(const double *data, const int N, const void *argv, double *result){
//    return XTRACT_SUCCESS;
//}

int xtract_subbands(const double *data, const int N, const void *argv, double *result)
{

    int n, bw, xtract_func, nbands, scale, start, lower, *argi, rv;

    argi = (int *)argv;

    xtract_func = argi[0];
    nbands = argi[1];
    scale = argi[2];
    start = argi[3];

    if(scale == XTRACT_LINEAR_SUBBANDS)
        bw = floorf((N - start) / nbands);
    else
        bw = start;

    lower = start;
    rv = XTRACT_SUCCESS;

    for(n = 0; n < nbands; n++)
    {

        /* Bounds sanity check */
        if(lower >= N || lower + bw >= N)
        {
            //   printf("n: %d\n", n);
            result[n] = 0.0;
            continue;
        }

        rv = xtract[xtract_func](data+lower, bw, NULL, &result[n]);

        if(rv != XTRACT_SUCCESS)
            return rv;

        switch(scale)
        {
        case XTRACT_OCTAVE_SUBBANDS:
            lower += bw;
            bw = lower;
            break;
        case XTRACT_LINEAR_SUBBANDS:
            lower += bw;
            break;
        }

    }

    return rv;

}