/tags/sphinxbase-0.7/src/sphinx_adtools/cont_fileseg.c

/* -*- c-basic-offset: 4; indent-tabs-mode: nil -*- */
/* ====================================================================
 * Copyright (c) 1999-2001 Carnegie Mellon University.  All rights
 * reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer. 
 *
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in
 *    the documentation and/or other materials provided with the
 *    distribution.
 *
 * This work was supported in part by funding from the Defense Advanced 
 * Research Projects Agency and the National Science Foundation of the 
 * United States of America, and the CMU Sphinx Speech Consortium.
 *
 * THIS SOFTWARE IS PROVIDED BY CARNEGIE MELLON UNIVERSITY ``AS IS'' AND 
 * ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, 
 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY
 * NOR ITS EMPLOYEES BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 * ====================================================================
 *
 */
/*
 * cont_fileseg.c -- Read input file, filter silence regions, and segment into utterances.
 * 
 * HISTORY
 * 
 * $Log: cont_fileseg.c,v $
 * Revision 1.1.1.1  2006/05/23 18:45:02  dhuggins
 * re-importation
 *
 * Revision 1.13  2005/06/30 00:28:46  rkm
 * Kept within-utterance silences in rawmode
 *
 * 
 * 28-Jun-2005	M K Ravishankar (rkm@cs.cmu.edu) at Carnegie Mellon University
 * 		Modified to use new state variables in cont_ad_t.
 * 
 * Revision 1.12  2005/05/31 15:54:38  rkm
 * *** empty log message ***
 *
 * Revision 1.11  2005/05/24 20:56:58  rkm
 * Added min/max-noise parameters to cont_fileseg
 *
 * Revision 1.10  2005/05/13 23:28:43  egouvea
 * Changed null device to system dependent one: NUL for windows, /dev/null for everything else
 * 
 * $Log: cont_fileseg.c,v $
 * Revision 1.1.1.1  2006/05/23 18:45:02  dhuggins
 * re-importation
 *
 * Revision 1.13  2005/06/30 00:28:46  rkm
 * Kept within-utterance silences in rawmode
 *
 * Revision 1.12  2005/05/31 15:54:38  rkm
 * *** empty log message ***
 *
 * Revision 1.11  2005/05/24 20:56:58  rkm
 * Added min/max-noise parameters to cont_fileseg
 *
 * Revision 1.9  2005/02/13 01:29:48  rkm
 * Fixed cont_ad_read to never cross sil/speech boundary, and rawmode
 *
 * Revision 1.8  2005/02/01 22:21:13  rkm
 * Added raw data logging, and raw data pass-through mode to cont_ad
 *
 * Revision 1.7  2004/07/16 00:57:11  egouvea
 * Added Ravi's implementation of FSG support.
 *
 * Revision 1.3  2004/06/25 14:58:05  rkm
 * *** empty log message ***
 *
 * Revision 1.2  2004/06/23 20:32:08  rkm
 * Exposed several cont_ad config parameters
 *
 * 
 * 27-Jun-96	M K Ravishankar (rkm@cs.cmu.edu) at Carnegie Mellon University
 * 		Created.
 */

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

#include <sphinxbase/prim_type.h>
#include <sphinxbase/ad.h>
#include <sphinxbase/cont_ad.h>
#include <sphinxbase/err.h>

static FILE *infp;              /* File being segmented */
static int32 swap;

/* Max size read by file_ad_read function on each invocation, for debugging */
static int32 max_ad_read_size;

#if defined(WIN32) && !defined(GNUWINCE)
#define NULL_DEVICE "NUL"
#else
#define NULL_DEVICE "/dev/null"
#endif


/*
 * Need to provide cont_ad_init with a read function to read the input file.
 * This is it.  The ad_rec_t *r argument is ignored since there is no A/D
 * device involved.
 */
static int32
file_ad_read(ad_rec_t * r, int16 * buf, int32 max)
{
    int32 i, k;

    if (max > max_ad_read_size)
        max = max_ad_read_size;

    k = fread(buf, sizeof(int16), max, infp);
    if (swap) {
        for (i = 0; i < k; i++) {
            buf[i] = ((buf[i] >> 8) & 0x00ff) | ((buf[i] << 8) & 0xff00);
        }
    }

    return ((k > 0) ? k : -1);
}


static void
usagemsg(char *pgm)
{
    E_INFO("Usage: %s \\\n", pgm);
    E_INFOCONT("\t[-? | -h] \\\n");
    E_INFOCONT("\t[-d | -debug] \\\n");
    E_INFOCONT("\t[-sps <sampling-rate> (16000)] \\\n");
    E_INFOCONT("\t[-b | -byteswap] \\\n");
    E_INFOCONT
        ("\t[{-s | -silsep} <length-silence-separator(sec) (0.5)]> \\\n");
    E_INFOCONT("\t[-w | -writeseg] \\\n");
    E_INFOCONT("\t[-min-noise <min-noise>] \\\n");
    E_INFOCONT("\t[-max-noise <max-noise>] \\\n");
    E_INFOCONT("\t[-delta-sil <delta-sil>] \\\n");
    E_INFOCONT("\t[-delta-speech <delta-speech>] \\\n");
    E_INFOCONT("\t[-sil-onset <sil-onset>] \\\n");
    E_INFOCONT("\t[-speech-onset <speech-onset>] \\\n");
    E_INFOCONT("\t[-adapt-rate <adapt-rate>] \\\n");
    E_INFOCONT("\t[-max-adreadsize <ad_read_blksize>] \\\n");
    E_INFOCONT("\t[-c <copy-input-file>] \\\n");
    E_INFOCONT("\t[-r | -rawmode] \\\n");
    E_INFOCONT("\t-i <input-file>\n");

    exit(0);
}

/*
 * Read specified input file, segment it into utterances wherever a silence segment of
 * a given minimum duration is encountered.  Filter out long silences.
 * Utterances are written to files named 00000000.raw, 00000001.raw, 00000002.raw, etc.
 */
int
main(int32 argc, char **argv)
{
    cont_ad_t *cont;
    int32 uttid, uttlen, starttime, siltime, sps, debug, writeseg, rawmode;
    int16 buf[4096];
    char *infile, *copyfile, segfile[1024];
    FILE *fp;
    float endsil;
    ad_rec_t ad;
    int32 i, k;
    int32 winsize, leader, trailer;
    int32 orig_min_noise, orig_max_noise;
    int32 orig_delta_sil, orig_delta_speech;
    int32 orig_speech_onset, orig_sil_onset;
    int32 min_noise, max_noise;
    int32 delta_sil, delta_speech;
    int32 sil_onset, speech_onset;
    float32 orig_adapt_rate;
    float32 adapt_rate;
    int32 total_speech_samples;
    float32 total_speech_sec;
    FILE *rawfp;

    /* Set argument defaults */
    cont = NULL;
    sps = 16000;
    swap = 0;
    endsil = 0.5;
    writeseg = 0;
    min_noise = max_noise = -1;
    delta_sil = delta_speech = -1;
    sil_onset = speech_onset = -1;
    adapt_rate = -1.0;
    max_ad_read_size = (int32) 0x7ffffff0;
    debug = 0;
    infile = NULL;
    copyfile = NULL;
    rawfp = NULL;
    rawmode = 0;

    /* Parse arguments */
    for (i = 1; i < argc; i++) {
        if ((strcmp(argv[i], "-help") == 0)
            || (strcmp(argv[i], "-h") == 0)
            || (strcmp(argv[i], "-?") == 0)) {
            usagemsg(argv[0]);
        }
        else if ((strcmp(argv[i], "-debug") == 0)
                 || (strcmp(argv[i], "-d") == 0)) {
            debug = 1;
        }
        else if (strcmp(argv[i], "-sps") == 0) {
            i++;
            if ((i == argc)
                || (sscanf(argv[i], "%d", &sps) != 1)
                || (sps <= 0)) {
                E_ERROR("Invalid -sps argument\n");
                usagemsg(argv[0]);
            }
        }
        else if ((strcmp(argv[i], "-byteswap") == 0)
                 || (strcmp(argv[i], "-b") == 0)) {
            swap = 1;
        }
        else if ((strcmp(argv[i], "-silsep") == 0)
                 || (strcmp(argv[i], "-s") == 0)) {
            i++;
            if ((i == argc)
                || (sscanf(argv[i], "%f", &endsil) != 1)
                || (endsil <= 0.0)) {
                E_ERROR("Invalid -silsep argument\n");
                usagemsg(argv[0]);
            }
        }
        else if ((strcmp(argv[i], "-writeseg") == 0)
                 || (strcmp(argv[i], "-w") == 0)) {
            writeseg = 1;
        }
        else if (strcmp(argv[i], "-min-noise") == 0) {
            i++;
            if ((i == argc) ||
                (sscanf(argv[i], "%d", &min_noise) != 1) ||
                (min_noise < 0)) {
                E_ERROR("Invalid -min-noise argument\n");
                usagemsg(argv[0]);
            }
        }
        else if (strcmp(argv[i], "-max-noise") == 0) {
            i++;
            if ((i == argc) ||
                (sscanf(argv[i], "%d", &max_noise) != 1) ||
                (max_noise < 0)) {
                E_ERROR("Invalid -max-noise argument\n");
                usagemsg(argv[0]);
            }
        }
        else if (strcmp(argv[i], "-delta-sil") == 0) {
            i++;
            if ((i == argc) ||
                (sscanf(argv[i], "%d", &delta_sil) != 1) ||
                (delta_sil < 0)) {
                E_ERROR("Invalid -delta-sil argument\n");
                usagemsg(argv[0]);
            }
        }
        else if (strcmp(argv[i], "-delta-speech") == 0) {
            i++;
            if ((i == argc) ||
                (sscanf(argv[i], "%d", &delta_speech) != 1) ||
                (delta_speech < 0)) {
                E_ERROR("Invalid -delta-speech argument\n");
                usagemsg(argv[0]);
            }
        }
        else if (strcmp(argv[i], "-sil-onset") == 0) {
            i++;
            if ((i == argc) ||
                (sscanf(argv[i], "%d", &sil_onset) != 1) ||
                (sil_onset < 1)) {
                E_ERROR("Invalid -sil-onset argument\n");
                usagemsg(argv[0]);
            }
        }
        else if (strcmp(argv[i], "-speech-onset") == 0) {
            i++;
            if ((i == argc) ||
                (sscanf(argv[i], "%d", &speech_onset) != 1) ||
                (speech_onset < 1)) {
                E_ERROR("Invalid -speech-onset argument\n");
                usagemsg(argv[0]);
            }
        }
        else if (strcmp(argv[i], "-adapt-rate") == 0) {
            i++;
            if ((i == argc) ||
                (sscanf(argv[i], "%f", &adapt_rate) != 1) ||
                (adapt_rate < 0.0) || (adapt_rate > 1.0)) {
                E_ERROR("Invalid -adapt-rate argument\n");
                usagemsg(argv[0]);
            }
        }
        else if (strcmp(argv[i], "-max-adreadsize") == 0) {
            i++;
            if ((i == argc) ||
                (sscanf(argv[i], "%d", &max_ad_read_size) != 1) ||
                (max_ad_read_size < 1)) {
                E_ERROR("Invalid -max-adreadsize argument\n");
                usagemsg(argv[0]);
            }
        }
        else if (strcmp(argv[i], "-c") == 0) {
            i++;
            if (i == argc) {
                E_ERROR("Invalid -c argument\n");
                usagemsg(argv[0]);
            }
            copyfile = argv[i];
        }
        else if ((strcmp(argv[i], "-rawmode") == 0)
                 || (strcmp(argv[i], "-r") == 0)) {
            rawmode = 1;
        }
        else if (strcmp(argv[i], "-i") == 0) {
            i++;
            if (i == argc) {
                E_ERROR("Invalid -i argument\n");
                usagemsg(argv[0]);
            }
            infile = argv[i];
        }
        else {
            usagemsg(argv[0]);
        }
    }

    if (infile == NULL) {
        E_ERROR("No input file specified\n");
        usagemsg(argv[0]);
    }

    if ((infp = fopen(infile, "rb")) == NULL)
        E_FATAL("Failed to open '%s' for reading: %s\n", infile, strerror(errno));

    /*
     * Associate continuous listening module with opened input file and read function.
     * No A/D device is involved, but need to fill in ad->sps.
     * Calibrate input data using first few seconds of file, but then rewind it!!
     */
    ad.sps = sps;
    ad.bps = sizeof(int16);
    if (!rawmode)
        cont = cont_ad_init(&ad, file_ad_read);
    else
        cont = cont_ad_init_rawmode(&ad, file_ad_read);

    printf("Calibrating ...");
    fflush(stdout);
    if (cont_ad_calib(cont) < 0)
        printf(" failed; file too short?\n");
    else
        printf(" done\n");
    rewind(infp);

    /* Convert desired min. inter-utterance silence duration to #samples */
    siltime = (int32) (endsil * sps);

    /* Enable writing raw input to output by the cont module if specified */
    if (copyfile) {
        if ((rawfp = fopen(copyfile, "wb")) == NULL)
            E_ERROR("Failed to open raw output file '%s' for writing: %s\n",
                    copyfile, strerror(errno));
        else
            cont_ad_set_rawfp(cont, rawfp);
    }

    cont_ad_get_params(cont,
                       &orig_delta_sil, &orig_delta_speech,
                       &orig_min_noise, &orig_max_noise,
                       &winsize,
                       &orig_speech_onset, &orig_sil_onset,
                       &leader, &trailer, &orig_adapt_rate);

    E_INFO("Default parameters:\n");
    E_INFOCONT("\tmin-noise = %d, max-noise = %d\n",
               orig_min_noise, orig_max_noise);
    E_INFOCONT("\tdelta-sil = %d, delta-speech = %d\n",
               orig_delta_sil, orig_delta_speech);
    E_INFOCONT("\tsil-onset = %d, speech-onset = %d\n",
               orig_sil_onset, orig_speech_onset);
    E_INFOCONT("\tadapt_rate = %.3f\n", orig_adapt_rate);

    if (min_noise < 0)
        min_noise = orig_min_noise;
    if (max_noise < 0)
        max_noise = orig_max_noise;
    if (delta_sil < 0)
        delta_sil = orig_delta_sil;
    if (delta_speech < 0)
        delta_speech = orig_delta_speech;
    if (sil_onset < 0)
        sil_onset = orig_sil_onset;
    if (speech_onset < 0)
        speech_onset = orig_speech_onset;
    if (adapt_rate < 0.0)
        adapt_rate = orig_adapt_rate;

    cont_ad_set_params(cont,
                       delta_sil, delta_speech,
                       min_noise, max_noise,
                       winsize,
                       speech_onset, sil_onset,
                       leader, trailer, adapt_rate);

    E_INFO("Current parameters:\n");
    E_INFOCONT("\tmin-noise = %d, max-noise = %d\n", min_noise, max_noise);
    E_INFOCONT("\tdelta-sil = %d, delta-speech = %d\n", delta_sil,
               delta_speech);
    E_INFOCONT("\tsil-onset = %d, speech-onset = %d\n", sil_onset,
               speech_onset);
    E_INFOCONT("\tadapt_rate = %.3f\n", adapt_rate);

    E_INFO("Sampling rate: %d", sps);
    E_INFOCONT("; Byteswap: %s", swap ? "Yes" : "No");
    E_INFOCONT("; Max ad-read size: %d\n", max_ad_read_size);

    if (debug)
        cont_ad_set_logfp(cont, stdout);

    total_speech_samples = 0;
    total_speech_sec = 0.0;

    uttid = 0;
    uttlen = 0;
    starttime = 0;
    fp = NULL;

    /* Process data */
    for (;;) {
        /* Get audio data from continuous listening module */
        k = cont_ad_read(cont, buf, 4096);

        if (k < 0) {            /* End of input audio file; close any open output file and exit */
            if (fp != NULL) {
                fclose(fp);
                fp = NULL;

                printf
                    ("Utt %08d, st= %8.2fs, et= %8.2fs, seg= %7.2fs (#samp= %10d)\n",
                     uttid, (double) starttime / (double) sps,
                     (double) (starttime + uttlen) / (double) sps,
                     (double) uttlen / (double) sps, uttlen);
                fflush(stdout);

                total_speech_samples += uttlen;
                total_speech_sec += (double) uttlen / (double) sps;

                uttid++;
            }

            break;
        }

        if (cont->state == CONT_AD_STATE_SIL) { /* Silence data got */
            if (fp != NULL) {   /* Currently in an utterance */
                if (cont->seglen > siltime) {   /* Long enough silence detected; end the utterance */
                    fclose(fp);
                    fp = NULL;

                    printf
                        ("Utt %08d, st= %8.2fs, et= %8.2fs, seg= %7.2fs (#samp= %10d)\n",
                         uttid, (double) starttime / (double) sps,
                         (double) (starttime + uttlen) / (double) sps,
                         (double) uttlen / (double) sps, uttlen);
                    fflush(stdout);

                    total_speech_samples += uttlen;
                    total_speech_sec += (double) uttlen / (double) sps;

                    uttid++;
                }
                else {
                    /*
                     * Short silence within utt; write it to output.  (Some extra trailing silence
                     * is included in the utterance, as a result.  Not to worry about it.)
                     */
                    if (k > 0) {
                        fwrite(buf, sizeof(int16), k, fp);
                        uttlen += k;
                    }
                }
            }
        }
        else {
            assert(cont->state == CONT_AD_STATE_SPEECH);

            if (fp == NULL) {   /* Not in an utt; open a new output file */
                if (writeseg)
                    sprintf(segfile, "%08d.raw", uttid);
                else
                    strcpy(segfile, NULL_DEVICE);
                if ((fp = fopen(segfile, "wb")) == NULL)
                    E_FATAL("Failed to open segmentation file '%s' for writing: %s\n", segfile, strerror(errno));

                starttime = cont->read_ts - k;
                uttlen = 0;
            }

            /* Write data obtained to output file */
            if (k > 0) {
                fwrite(buf, sizeof(int16), k, fp);
                uttlen += k;
            }
        }
    }

    if (rawfp)
        fclose(rawfp);

    E_INFO("Total raw input speech = %d frames, %d samples, %.2f sec\n",
           cont->tot_frm, cont->tot_frm * cont->spf,
           (cont->tot_frm * cont->spf) / (float32) cont->sps);
    E_INFO("Total speech detected = %d samples, %.2f sec\n",
           total_speech_samples, total_speech_sec);

    cont_ad_close(cont);

    return 0;
}