src/libespeak-ng/wavegen.c

/*
 * Copyright (C) 2005 to 2013 by Jonathan Duddington
 * email: jonsd@users.sourceforge.net
 * Copyright (C) 2015-2016 Reece H. Dunn
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 3 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, see: <http://www.gnu.org/licenses/>.
 */

// this version keeps wavemult window as a constant fraction
// of the cycle length - but that spreads out the HF peaks too much

#include "config.h"

#include <math.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include <espeak-ng/espeak_ng.h>
#include <espeak-ng/speak_lib.h>

#include "wavegen.h"
#include "common.h"                   // for espeak_rand
#include "synthesize.h"               // for WGEN_DATA, RESONATOR, frame_t
#include "mbrola.h"                  // for MbrolaFill, MbrolaReset, mbrola...

#if USE_KLATT
#include "klatt.h"
#endif

#if USE_LIBSONIC
#include "sonic.h"
#endif

#include "sintab.h"
#include "speech.h"

static void SetSynth(int length, int modn, frame_t *fr1, frame_t *fr2, voice_t *v);

static voice_t *wvoice = NULL;

static int option_harmonic1 = 10;
static int flutter_amp = 64;

static int general_amplitude = 60;
static int consonant_amp = 26;

int embedded_value[N_EMBEDDED_VALUES];

static int PHASE_INC_FACTOR;
int samplerate = 0; // this is set by Wavegeninit()

static wavegen_peaks_t peaks[N_PEAKS];
static int peak_harmonic[N_PEAKS];
static int peak_height[N_PEAKS];

int echo_head;
int echo_tail;
int echo_amp = 0;
short echo_buf[N_ECHO_BUF];
static int echo_length = 0; // period (in sample\) to ensure completion of echo at the end of speech, set in WavegenSetEcho()

static int voicing;
static RESONATOR rbreath[N_PEAKS];

#define N_LOWHARM  30
#define MAX_HARMONIC 400 // 400 * 50Hz = 20 kHz, more than enough
static int harm_inc[N_LOWHARM]; // only for these harmonics do we interpolate amplitude between steps
static int *harmspect;
static int hswitch = 0;
static int hspect[2][MAX_HARMONIC]; // 2 copies, we interpolate between then

static int nsamples = 0; // number to do
static int modulation_type = 0;
static int glottal_flag = 0;
static int glottal_reduce = 0;

static WGEN_DATA wdata;

static int amp_ix;
static int amp_inc;
static unsigned char *amplitude_env = NULL;

static int samplecount = 0; // number done
static int samplecount_start = 0; // count at start of this segment
static int end_wave = 0; // continue to end of wave cycle
static int wavephase;
static int phaseinc;
static int cycle_samples; // number of samples in a cycle at current pitch
static int cbytes;
static int hf_factor;

static double minus_pi_t;
static double two_pi_t;

unsigned char *out_ptr;
unsigned char *out_end;

espeak_ng_OUTPUT_HOOKS* output_hooks = NULL;
static int const_f0 = 0;

// the queue of operations passed to wavegen from sythesize
intptr_t wcmdq[N_WCMDQ][4];
int wcmdq_head = 0;
int wcmdq_tail = 0;

// pitch,speed,
const int embedded_default[N_EMBEDDED_VALUES]    = { 0,     50, espeakRATE_NORMAL, 100, 50,  0,  0, 0, espeakRATE_NORMAL, 0, 0, 0, 0, 0, 0 };
static const int embedded_max[N_EMBEDDED_VALUES] = { 0, 0x7fff, 750, 300, 99, 99, 99, 0, 750, 0, 0, 0, 0, 4, 0 };

#if USE_LIBSONIC
static sonicStream sonicSpeedupStream = NULL;
static double sonicSpeed = 1.0;
#endif

// 1st index=roughness
// 2nd index=modulation_type
// value: bits 0-3  amplitude (16ths), bits 4-7 every n cycles
#define N_ROUGHNESS 8
static const unsigned char modulation_tab[N_ROUGHNESS][8] = {
	{ 0, 0x00, 0x00, 0x00, 0, 0x46, 0xf2, 0x29 },
	{ 0, 0x2f, 0x00, 0x2f, 0, 0x45, 0xf2, 0x29 },
	{ 0, 0x2f, 0x00, 0x2e, 0, 0x45, 0xf2, 0x28 },
	{ 0, 0x2e, 0x00, 0x2d, 0, 0x34, 0xf2, 0x28 },
	{ 0, 0x2d, 0x2d, 0x2c, 0, 0x34, 0xf2, 0x28 },
	{ 0, 0x2b, 0x2b, 0x2b, 0, 0x34, 0xf2, 0x28 },
	{ 0, 0x2a, 0x2a, 0x2a, 0, 0x34, 0xf2, 0x28 },
	{ 0, 0x29, 0x29, 0x29, 0, 0x34, 0xf2, 0x28 },
};

// Flutter table, to add natural variations to the pitch
#define N_FLUTTER  0x170
static int Flutter_inc;
static const unsigned char Flutter_tab[N_FLUTTER] = {
	0x80, 0x9b, 0xb5, 0xcb, 0xdc, 0xe8, 0xed, 0xec,
	0xe6, 0xdc, 0xce, 0xbf, 0xb0, 0xa3, 0x98, 0x90,
	0x8c, 0x8b, 0x8c, 0x8f, 0x92, 0x94, 0x95, 0x92,
	0x8c, 0x83, 0x78, 0x69, 0x59, 0x49, 0x3c, 0x31,
	0x2a, 0x29, 0x2d, 0x36, 0x44, 0x56, 0x69, 0x7d,
	0x8f, 0x9f, 0xaa, 0xb1, 0xb2, 0xad, 0xa4, 0x96,
	0x87, 0x78, 0x69, 0x5c, 0x53, 0x4f, 0x4f, 0x55,
	0x5e, 0x6b, 0x7a, 0x88, 0x96, 0xa2, 0xab, 0xb0,

	0xb1, 0xae, 0xa8, 0xa0, 0x98, 0x91, 0x8b, 0x88,
	0x89, 0x8d, 0x94, 0x9d, 0xa8, 0xb2, 0xbb, 0xc0,
	0xc1, 0xbd, 0xb4, 0xa5, 0x92, 0x7c, 0x63, 0x4a,
	0x32, 0x1e, 0x0e, 0x05, 0x02, 0x05, 0x0f, 0x1e,
	0x30, 0x44, 0x59, 0x6d, 0x7f, 0x8c, 0x96, 0x9c,
	0x9f, 0x9f, 0x9d, 0x9b, 0x99, 0x99, 0x9c, 0xa1,
	0xa9, 0xb3, 0xbf, 0xca, 0xd5, 0xdc, 0xe0, 0xde,
	0xd8, 0xcc, 0xbb, 0xa6, 0x8f, 0x77, 0x60, 0x4b,

	0x3a, 0x2e, 0x28, 0x29, 0x2f, 0x3a, 0x48, 0x59,
	0x6a, 0x7a, 0x86, 0x90, 0x94, 0x95, 0x91, 0x89,
	0x80, 0x75, 0x6b, 0x62, 0x5c, 0x5a, 0x5c, 0x61,
	0x69, 0x74, 0x80, 0x8a, 0x94, 0x9a, 0x9e, 0x9d,
	0x98, 0x90, 0x86, 0x7c, 0x71, 0x68, 0x62, 0x60,
	0x63, 0x6b, 0x78, 0x88, 0x9b, 0xaf, 0xc2, 0xd2,
	0xdf, 0xe6, 0xe7, 0xe2, 0xd7, 0xc6, 0xb2, 0x9c,
	0x84, 0x6f, 0x5b, 0x4b, 0x40, 0x39, 0x37, 0x38,

	0x3d, 0x43, 0x4a, 0x50, 0x54, 0x56, 0x55, 0x52,
	0x4d, 0x48, 0x42, 0x3f, 0x3e, 0x41, 0x49, 0x56,
	0x67, 0x7c, 0x93, 0xab, 0xc3, 0xd9, 0xea, 0xf6,
	0xfc, 0xfb, 0xf4, 0xe7, 0xd5, 0xc0, 0xaa, 0x94,
	0x80, 0x71, 0x64, 0x5d, 0x5a, 0x5c, 0x61, 0x68,
	0x70, 0x77, 0x7d, 0x7f, 0x7f, 0x7b, 0x74, 0x6b,
	0x61, 0x57, 0x4e, 0x48, 0x46, 0x48, 0x4e, 0x59,
	0x66, 0x75, 0x84, 0x93, 0x9f, 0xa7, 0xab, 0xaa,

	0xa4, 0x99, 0x8b, 0x7b, 0x6a, 0x5b, 0x4e, 0x46,
	0x43, 0x45, 0x4d, 0x5a, 0x6b, 0x7f, 0x92, 0xa6,
	0xb8, 0xc5, 0xcf, 0xd3, 0xd2, 0xcd, 0xc4, 0xb9,
	0xad, 0xa1, 0x96, 0x8e, 0x89, 0x87, 0x87, 0x8a,
	0x8d, 0x91, 0x92, 0x91, 0x8c, 0x84, 0x78, 0x68,
	0x55, 0x41, 0x2e, 0x1c, 0x0e, 0x05, 0x01, 0x05,
	0x0f, 0x1f, 0x34, 0x4d, 0x68, 0x81, 0x9a, 0xb0,
	0xc1, 0xcd, 0xd3, 0xd3, 0xd0, 0xc8, 0xbf, 0xb5,

	0xab, 0xa4, 0x9f, 0x9c, 0x9d, 0xa0, 0xa5, 0xaa,
	0xae, 0xb1, 0xb0, 0xab, 0xa3, 0x96, 0x87, 0x76,
	0x63, 0x51, 0x42, 0x36, 0x2f, 0x2d, 0x31, 0x3a,
	0x48, 0x59, 0x6b, 0x7e, 0x8e, 0x9c, 0xa6, 0xaa,
	0xa9, 0xa3, 0x98, 0x8a, 0x7b, 0x6c, 0x5d, 0x52,
	0x4a, 0x48, 0x4a, 0x50, 0x5a, 0x67, 0x75, 0x82
};

// waveform shape table for HF peaks, formants 6,7,8
#define N_WAVEMULT 128
static int wavemult_offset = 0;
static int wavemult_max = 0;

// the presets are for 22050 Hz sample rate.
// A different rate will need to recalculate the presets in WavegenInit()
static unsigned char wavemult[N_WAVEMULT] = {
	  0,   0,   0,   2,   3,   5,   8,  11,  14,  18,  22,  27,  32,  37,  43,  49,
	 55,  62,  69,  76,  83,  90,  98, 105, 113, 121, 128, 136, 144, 152, 159, 166,
	174, 181, 188, 194, 201, 207, 213, 218, 224, 228, 233, 237, 240, 244, 246, 249,
	251, 252, 253, 253, 253, 253, 252, 251, 249, 246, 244, 240, 237, 233, 228, 224,
	218, 213, 207, 201, 194, 188, 181, 174, 166, 159, 152, 144, 136, 128, 121, 113,
	105,  98,  90,  83,  76,  69,  62,  55,  49,  43,  37,  32,  27,  22,  18,  14,
	 11,   8,   5,   3,   2,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
	  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0
};

// set from y = pow(2,x) * 128,  x=-1 to 1
#define MAX_PITCH_VALUE  101
static const unsigned char pitch_adjust_tab[MAX_PITCH_VALUE+1] = {
	 64,  65,  66,  67,  68,  69,  70,  71,
	 72,  73,  74,  75,  76,  77,  78,  79,
	 80,  81,  82,  83,  84,  86,  87,  88,
	 89,  91,  92,  93,  94,  96,  97,  98,
	100, 101, 103, 104, 105, 107, 108, 110,
	111, 113, 115, 116, 118, 119, 121, 123,
	124, 126, 128, 130, 132, 133, 135, 137,
	139, 141, 143, 145, 147, 149, 151, 153,
	155, 158, 160, 162, 164, 167, 169, 171,
	174, 176, 179, 181, 184, 186, 189, 191,
	194, 197, 199, 202, 205, 208, 211, 214,
	217, 220, 223, 226, 229, 232, 236, 239,
	242, 246, 249, 252, 254, 255
};

void WcmdqStop(void)
{
	wcmdq_head = 0;
	wcmdq_tail = 0;

#if USE_LIBSONIC
	if (sonicSpeedupStream != NULL) {
		sonicDestroyStream(sonicSpeedupStream);
		sonicSpeedupStream = NULL;
	}
#endif

#if USE_MBROLA
	if (mbrola_name[0] != 0)
		MbrolaReset();
#endif
}

int WcmdqFree(void)
{
	int i;
	i = wcmdq_head - wcmdq_tail;
	if (i <= 0) i += N_WCMDQ;
	return i;
}

int WcmdqUsed(void)
{
	return N_WCMDQ - WcmdqFree();
}

void WcmdqInc(void)
{
	wcmdq_tail++;
	if (wcmdq_tail >= N_WCMDQ) wcmdq_tail = 0;
}

static void WcmdqIncHead(void)
{
	MAKE_MEM_UNDEFINED(&wcmdq[wcmdq_head], sizeof(wcmdq[wcmdq_head]));
	wcmdq_head++;
	if (wcmdq_head >= N_WCMDQ) wcmdq_head = 0;
}

#define PEAKSHAPEW 256

static const unsigned char pk_shape1[PEAKSHAPEW+1] = {
	255, 254, 254, 254, 254, 254, 253, 253, 252, 251, 251, 250, 249, 248, 247, 246,
	245, 244, 242, 241, 239, 238, 236, 234, 233, 231, 229, 227, 225, 223, 220, 218,
	216, 213, 211, 209, 207, 205, 203, 201, 199, 197, 195, 193, 191, 189, 187, 185,
	183, 180, 178, 176, 173, 171, 169, 166, 164, 161, 159, 156, 154, 151, 148, 146,
	143, 140, 138, 135, 132, 129, 126, 123, 120, 118, 115, 112, 108, 105, 102,  99,
	 96,  95,  93,  91,  90,  88,  86,  85,  83,  82,  80,  79,  77,  76,  74,  73,
	 72,  70,  69,  68,  67,  66,  64,  63,  62,  61,  60,  59,  58,  57,  56,  55,
	 55,  54,  53,  52,  52,  51,  50,  50,  49,  48,  48,  47,  47,  46,  46,  46,
	 45,  45,  45,  44,  44,  44,  44,  44,  44,  44,  43,  43,  43,  43,  44,  43,
	 42,  42,  41,  40,  40,  39,  38,  38,  37,  36,  36,  35,  35,  34,  33,  33,
	 32,  32,  31,  30,  30,  29,  29,  28,  28,  27,  26,  26,  25,  25,  24,  24,
	 23,  23,  22,  22,  21,  21,  20,  20,  19,  19,  18,  18,  18,  17,  17,  16,
	 16,  15,  15,  15,  14,  14,  13,  13,  13,  12,  12,  11,  11,  11,  10,  10,
	 10,   9,   9,   9,   8,   8,   8,   7,   7,   7,   7,   6,   6,   6,   5,   5,
	  5,   5,   4,   4,   4,   4,   4,   3,   3,   3,   3,   2,   2,   2,   2,   2,
	  2,   1,   1,   1,   1,   1,   1,   0,   0,   0,   0,   0,   0,   0,   0,   0,
	  0
};

static const unsigned char pk_shape2[PEAKSHAPEW+1] = {
	255, 254, 254, 254, 254, 254, 254, 254, 254, 254, 253, 253, 253, 253, 252, 252,
	252, 251, 251, 251, 250, 250, 249, 249, 248, 248, 247, 247, 246, 245, 245, 244,
	243, 243, 242, 241, 239, 237, 235, 233, 231, 229, 227, 225, 223, 221, 218, 216,
	213, 211, 208, 205, 203, 200, 197, 194, 191, 187, 184, 181, 178, 174, 171, 167,
	163, 160, 156, 152, 148, 144, 140, 136, 132, 127, 123, 119, 114, 110, 105, 100,
	 96,  94,  91,  88,  86,  83,  81,  78,  76,  74,  71,  69,  66,  64,  62,  60,
	 57,  55,  53,  51,  49,  47,  44,  42,  40,  38,  36,  34,  32,  30,  29,  27,
	 25,  23,  21,  19,  18,  16,  14,  12,  11,   9,   7,   6,   4,   3,   1,   0,
	  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
	  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
	  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
	  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
	  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
	  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
	  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
	  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
	  0
};

static const unsigned char *pk_shape;

void WavegenInit(int rate, int wavemult_fact)
{
	int ix;
	double x;

	if (wavemult_fact == 0)
		wavemult_fact = 60; // default

	wvoice = NULL;
	samplerate = rate;
	PHASE_INC_FACTOR = 0x8000000 / samplerate; // assumes pitch is Hz*32
	Flutter_inc = (64 * samplerate)/rate;
	samplecount = 0;
	nsamples = 0;
	wavephase = 0x7fffffff;

	wdata.amplitude = 32;
	wdata.amplitude_fmt = 100;

	for (ix = 0; ix < N_EMBEDDED_VALUES; ix++)
		embedded_value[ix] = embedded_default[ix];

	// set up window to generate a spread of harmonics from a
	// single peak for HF peaks
	wavemult_max = (samplerate * wavemult_fact)/(256 * 50);
	if (wavemult_max > N_WAVEMULT) wavemult_max = N_WAVEMULT;

	wavemult_offset = wavemult_max/2;

	if (samplerate != 22050) {
		// wavemult table has preset values for 22050 Hz, we only need to
		// recalculate them if we have a different sample rate
		for (ix = 0; ix < wavemult_max; ix++) {
			x = 127*(1.0 - cos((M_PI*2)*ix/wavemult_max));
			wavemult[ix] = (int)x;
		}
	}

	pk_shape = pk_shape2;

#if USE_KLATT
	KlattInit();
#endif
}

void WavegenFini(void)
{
#if USE_KLATT
	KlattFini();
#endif
}

int GetAmplitude(void)
{
	int amp;

	// normal, none, reduced, moderate, strong
	static const unsigned char amp_emphasis[5] = { 16, 16, 10, 16, 22 };

	amp = (embedded_value[EMBED_A])*55/100;
	general_amplitude = amp * amp_emphasis[embedded_value[EMBED_F]] / 16;
	return general_amplitude;
}

static void WavegenSetEcho(void)
{
	if (wvoice == NULL)
		return;

	int delay;
	int amp;

	voicing = wvoice->voicing;
	delay = wvoice->echo_delay;
	amp = wvoice->echo_amp;

	if (delay >= N_ECHO_BUF)
		delay = N_ECHO_BUF-1;
	if (amp > 100)
		amp = 100;

	memset(echo_buf, 0, sizeof(echo_buf));
	echo_tail = 0;

	if (embedded_value[EMBED_H] > 0) {
		// set echo from an embedded command in the text
		amp = embedded_value[EMBED_H];
		delay = 130;
	}

	if (delay == 0)
		amp = 0;

	echo_head = (delay * samplerate)/1000;
	echo_length = echo_head; // ensure completion of echo at the end of speech. Use 1 delay period?
	if (amp == 0)
		echo_length = 0;
	if (amp > 20)
		echo_length = echo_head * 2; // perhaps allow 2 echo periods if the echo is loud.

	// echo_amp units are 1/256ths of the amplitude of the original sound.
	echo_amp = amp;
	// compensate (partially) for increase in amplitude due to echo
	general_amplitude = GetAmplitude();
	general_amplitude = ((general_amplitude * (500-amp))/500);
}

int PeaksToHarmspect(wavegen_peaks_t *peaks, int pitch, int *htab, int control)
{
	if (wvoice == NULL)
		return 1;

	// Calculate the amplitude of each  harmonics from the formants
	// Only for formants 0 to 5

	// control 0=initial call, 1=every 64 cycles

	// pitch and freqs are Hz<<16

	int f;
	wavegen_peaks_t *p;
	int fp;  // centre freq of peak
	int fhi; // high freq of peak
	int h;   // harmonic number
	int pk;
	int hmax;
	int hmax_samplerate; // highest harmonic allowed for the samplerate
	int x;
	int h1;

	// initialise as much of *out as we will need
	hmax = (peaks[wvoice->n_harmonic_peaks].freq + peaks[wvoice->n_harmonic_peaks].right)/pitch;
	if (hmax >= MAX_HARMONIC)
		hmax = MAX_HARMONIC-1;

	// restrict highest harmonic to half the samplerate
	hmax_samplerate = (((samplerate * 19)/40) << 16)/pitch; // only 95% of Nyquist freq

	if (hmax > hmax_samplerate)
		hmax = hmax_samplerate;

	for (h = 0; h <= hmax; h++)
		htab[h] = 0;

	for (pk = 0; pk <= wvoice->n_harmonic_peaks; pk++) {
		p = &peaks[pk];
		if ((p->height == 0) || (fp = p->freq) == 0)
			continue;

		fhi = p->freq + p->right;
		h = ((p->freq - p->left) / pitch) + 1;
		if (h <= 0) h = 1;

		for (f = pitch*h; f < fp; f += pitch)
			htab[h++] += pk_shape[(fp-f)/(p->left>>8)] * p->height;
		for (; f < fhi; f += pitch)
			htab[h++] += pk_shape[(f-fp)/(p->right>>8)] * p->height;
	}

	int y;
	int h2;
	// increase bass
	y = peaks[1].height * 10; // addition as a multiple of 1/256s
	h2 = (1000<<16)/pitch; // decrease until 1000Hz
	if (h2 > 0) {
		x = y/h2;
		h = 1;
		while (y > 0) {
			htab[h++] += y;
			y -= x;
		}
	}

	// find the nearest harmonic for HF peaks where we don't use shape
	for (; pk < N_PEAKS; pk++) {
		x = peaks[pk].height >> 14;
		peak_height[pk] = (x * x * 5)/2;

		// find the nearest harmonic for HF peaks where we don't use shape
		if (control == 0) {
			// set this initially, but make changes only at the quiet point
			peak_harmonic[pk] = peaks[pk].freq / pitch;
		}
		// only use harmonics up to half the samplerate
		if (peak_harmonic[pk] >= hmax_samplerate)
			peak_height[pk] = 0;
	}

	// convert from the square-rooted values
	f = 0;
	for (h = 0; h <= hmax; h++, f += pitch) {
		x = htab[h] >> 15;
		htab[h] = (x * x) >> 8;

		int ix;
		if ((ix = (f >> 19)) < N_TONE_ADJUST)
			htab[h] = (htab[h] * wvoice->tone_adjust[ix]) >> 13; // index tone_adjust with Hz/8
	}

	// adjust the amplitude of the first harmonic, affects tonal quality
	h1 = htab[1] * option_harmonic1;
	htab[1] = h1/8;

	// calc intermediate increments of LF harmonics
	if (control & 1) {
		for (h = 1; h < N_LOWHARM; h++)
			harm_inc[h] = (htab[h] - harmspect[h]) >> 3;
	}

	return hmax; // highest harmonic number
}

static void AdvanceParameters(void)
{
	// Called every 64 samples to increment the formant freq, height, and widths
	if (wvoice == NULL)
		return;

	int x;
	int ix;
	static int Flutter_ix = 0;

	// advance the pitch
	wdata.pitch_ix += wdata.pitch_inc;
	if ((ix = wdata.pitch_ix>>8) > 127) ix = 127;
	x = wdata.pitch_env[ix] * wdata.pitch_range;
	wdata.pitch = (x>>8) + wdata.pitch_base;
	
	

	amp_ix += amp_inc;

	/* add pitch flutter */
	if (Flutter_ix >= (N_FLUTTER*64))
		Flutter_ix = 0;
	x = ((int)(Flutter_tab[Flutter_ix >> 6])-0x80) * flutter_amp;
	Flutter_ix += Flutter_inc;
	wdata.pitch += x;
	
	if(const_f0)
		wdata.pitch = (const_f0<<12);
	
	if (wdata.pitch < 102400)
		wdata.pitch = 102400; // min pitch, 25 Hz  (25 << 12)

	if (samplecount == samplecount_start)
		return;

	for (ix = 0; ix <= wvoice->n_harmonic_peaks; ix++) {
		peaks[ix].freq1 += peaks[ix].freq_inc;
		peaks[ix].freq = (int)peaks[ix].freq1;
		peaks[ix].height1 += peaks[ix].height_inc;
		if ((peaks[ix].height = (int)peaks[ix].height1) < 0)
			peaks[ix].height = 0;
		peaks[ix].left1 += peaks[ix].left_inc;
		peaks[ix].left = (int)peaks[ix].left1;
		if (ix < 3) {
			peaks[ix].right1 += peaks[ix].right_inc;
			peaks[ix].right = (int)peaks[ix].right1;
		} else
			peaks[ix].right = peaks[ix].left;
	}
	for (; ix < 8; ix++) {
		// formants 6,7,8 don't have a width parameter
		if (ix < 7) {
			peaks[ix].freq1 += peaks[ix].freq_inc;
			peaks[ix].freq = (int)peaks[ix].freq1;
		}
		peaks[ix].height1 += peaks[ix].height_inc;
		if ((peaks[ix].height = (int)peaks[ix].height1) < 0)
			peaks[ix].height = 0;
	}
}

static double resonator(RESONATOR *r, double input)
{
	double x;

	x = r->a * input + r->b * r->x1 + r->c * r->x2;
	r->x2 = r->x1;
	r->x1 = x;

	return x;
}

static void setresonator(RESONATOR *rp, int freq, int bwidth, int init)
{
	// freq    Frequency of resonator in Hz
	// bwidth  Bandwidth of resonator in Hz
	// init    Initialize internal data

	double x;
	double arg;

	if (init) {
		rp->x1 = 0;
		rp->x2 = 0;
	}

	arg = minus_pi_t * bwidth;
	x = exp(arg);

	rp->c = -(x * x);

	arg = two_pi_t * freq;
	rp->b = x * cos(arg) * 2.0;

	rp->a = 1.0 - rp->b - rp->c;
}

void InitBreath(void)
{
	int ix;

	minus_pi_t = -M_PI / samplerate;
	two_pi_t = -2.0 * minus_pi_t;

	for (ix = 0; ix < N_PEAKS; ix++)
		setresonator(&rbreath[ix], 2000, 200, 1);
}

static void SetBreath(void)
{
	int pk;

	if (wvoice == NULL || wvoice->breath[0] == 0)
		return;

	for (pk = 1; pk < N_PEAKS; pk++) {
		if (wvoice->breath[pk] != 0) {
			// breath[0] indicates that some breath formants are needed
			// set the freq from the current synthesis formant and the width from the voice data
			setresonator(&rbreath[pk], peaks[pk].freq >> 16, wvoice->breathw[pk], 0);
		}
	}
}

static int ApplyBreath(void)
{
	if (wvoice == NULL)
		return 0;

	int value = 0;
	int noise;
	int ix;

	// use two random numbers, for alternate formants
	noise = espeak_rand(-0x2000, 0x1fff);

	for (ix = 1; ix < N_PEAKS; ix++) {
		int amp;
		if ((amp = wvoice->breath[ix]) != 0) {
			amp *= (peaks[ix].height >> 14);
			value += (int)resonator(&rbreath[ix], noise) * amp;
		}
	}
	return value;
}

static int Wavegen(int length, int modulation, bool resume, frame_t *fr1, frame_t *fr2, voice_t *wvoice)
{
	if (resume == false)
		SetSynth(length, modulation, fr1, fr2, wvoice);

	if (wvoice == NULL)
		return 0;

	unsigned short waveph;
	unsigned short theta;
	int total;
	int h;
	int ix;
	int z, z1, z2;
	int echo;
	int ov;
	static int maxh, maxh2;
	int pk;
	signed char c;
	int sample;
	int amp;
	int modn_amp = 1, modn_period;
	static int agc = 256;
	static int h_switch_sign = 0;
	static int cycle_count = 0;
	static int amplitude2 = 0; // adjusted for pitch

	// continue until the output buffer is full, or
	// the required number of samples have been produced

	for (;;) {
		if ((end_wave == 0) && (samplecount == nsamples))
			return 0;

		if ((samplecount & 0x3f) == 0) {
			// every 64 samples, adjust the parameters
			if (samplecount == 0) {
				hswitch = 0;
				harmspect = hspect[0];
				maxh2 = PeaksToHarmspect(peaks, wdata.pitch<<4, hspect[0], 0);

				// adjust amplitude to compensate for fewer harmonics at higher pitch
				amplitude2 = (wdata.amplitude * (wdata.pitch >> 8) * wdata.amplitude_fmt)/(10000 << 3);

				// switch sign of harmonics above about 900Hz, to reduce max peak amplitude
				h_switch_sign = 890 / (wdata.pitch >> 12);
			} else
				AdvanceParameters();

			// pitch is Hz<<12
			phaseinc = (wdata.pitch>>7) * PHASE_INC_FACTOR;
			cycle_samples = samplerate/(wdata.pitch >> 12); // sr/(pitch*2)
			hf_factor = wdata.pitch >> 11;

			maxh = maxh2;
			harmspect = hspect[hswitch];
			hswitch ^= 1;
			maxh2 = PeaksToHarmspect(peaks, wdata.pitch<<4, hspect[hswitch], 1);

			SetBreath();
		} else if ((samplecount & 0x07) == 0) {
			for (h = 1; h < N_LOWHARM && h <= maxh2 && h <= maxh; h++)
				harmspect[h] += harm_inc[h];

			// bring automatic gain control back towards unity
			if (agc < 256) agc++;
		}

		samplecount++;

		if (wavephase > 0) {
			wavephase += phaseinc;
			if (wavephase < 0) {
				// sign has changed, reached a quiet point in the waveform
				cbytes = wavemult_offset - (cycle_samples)/2;
				if (samplecount > nsamples)
					return 0;

				cycle_count++;

				for (pk = wvoice->n_harmonic_peaks+1; pk < N_PEAKS; pk++) {
					// find the nearest harmonic for HF peaks where we don't use shape
					peak_harmonic[pk] = ((peaks[pk].freq / (wdata.pitch*8)) + 1) / 2;
				}

				// adjust amplitude to compensate for fewer harmonics at higher pitch
				amplitude2 = (wdata.amplitude * (wdata.pitch >> 8) * wdata.amplitude_fmt)/(10000 << 3);

				if (glottal_flag > 0) {
					if (glottal_flag == 3) {
						if ((nsamples-samplecount) < (cycle_samples*2)) {
							// Vowel before glottal-stop.
							// This is the start of the penultimate cycle, reduce its amplitude
							glottal_flag = 2;
							amplitude2 = (amplitude2 *  glottal_reduce)/256;
						}
					} else if (glottal_flag == 4) {
						// Vowel following a glottal-stop.
						// This is the start of the second cycle, reduce its amplitude
						glottal_flag = 2;
						amplitude2 = (amplitude2 * glottal_reduce)/256;
					} else
						glottal_flag--;
				}

				if (amplitude_env != NULL) {
					// amplitude envelope is only used for creaky voice effect on certain vowels/tones
					if ((ix = amp_ix>>8) > 127) ix = 127;
					amp = amplitude_env[ix];
					amplitude2 = (amplitude2 * amp)/128;
				}

				// introduce roughness into the sound by reducing the amplitude of
				modn_period = 0;
				if (voice->roughness < N_ROUGHNESS) {
					modn_period = modulation_tab[voice->roughness][modulation_type];
					modn_amp = modn_period & 0xf;
					modn_period = modn_period >> 4;
				}

				if (modn_period != 0) {
					if (modn_period == 0xf) {
						// just once */
						amplitude2 = (amplitude2 * modn_amp)/16;
						modulation_type = 0;
					} else {
						// reduce amplitude every [modn_period} cycles
						if ((cycle_count % modn_period) == 0)
							amplitude2 = (amplitude2 * modn_amp)/16;
					}
				}
			}
		} else
			wavephase += phaseinc;
		waveph = (unsigned short)(wavephase >> 16);
		total = 0;

		// apply HF peaks, formants 6,7,8
		// add a single harmonic and then spread this my multiplying by a
		// window.  This is to reduce the processing power needed to add the
		// higher frequence harmonics.
		cbytes++;
		if (cbytes >= 0 && cbytes < wavemult_max) {
			for (pk = wvoice->n_harmonic_peaks+1; pk < N_PEAKS; pk++) {
				theta = peak_harmonic[pk] * waveph;
				total += (long)sin_tab[theta >> 5] * peak_height[pk];
			}

			// spread the peaks by multiplying by a window
			total = (long)(total / hf_factor) * wavemult[cbytes];
		}

		// apply main peaks, formants 0 to 5
		theta = waveph;

		for (h = 1; h <= h_switch_sign; h++) {
			total += ((int)sin_tab[theta >> 5] * harmspect[h]);
			theta += waveph;
		}
		while (h <= maxh) {
			total -= ((int)sin_tab[theta >> 5] * harmspect[h]);
			theta += waveph;
			h++;
		}

		if (voicing != 64)
			total = (total >> 6) * voicing;

		if (wvoice->breath[0])
			total +=  ApplyBreath();

		// mix with sampled wave if required
		z2 = 0;
		if (wdata.mix_wavefile_ix < wdata.n_mix_wavefile) {
			if (wdata.mix_wave_scale == 0) {
				// a 16 bit sample
				c = wdata.mix_wavefile[wdata.mix_wavefile_ix+wdata.mix_wavefile_offset+1];
				sample = wdata.mix_wavefile[wdata.mix_wavefile_ix+wdata.mix_wavefile_offset] + (c * 256);
				wdata.mix_wavefile_ix += 2;
			} else {
				// a 8 bit sample, scaled
				sample = (signed char)wdata.mix_wavefile[wdata.mix_wavefile_offset+wdata.mix_wavefile_ix++] * wdata.mix_wave_scale;
			}
			z2 = (sample * wdata.amplitude_v) >> 10;
			z2 = (z2 * wdata.mix_wave_amp)/32;

			if ((wdata.mix_wavefile_ix + wdata.mix_wavefile_offset) >= wdata.mix_wavefile_max)  // reached the end of available WAV data
				wdata.mix_wavefile_offset -= (wdata.mix_wavefile_max*3)/4;
		}

		z1 = z2 + (((total>>8) * amplitude2) >> 13);

		echo = (echo_buf[echo_tail++] * echo_amp);
		z1 += echo >> 8;
		if (echo_tail >= N_ECHO_BUF)
			echo_tail = 0;

		z = (z1 * agc) >> 8;

		// check for overflow, 16bit signed samples
		if (z >= 32768) {
			ov = 8388608/z1 - 1;      // 8388608 is 2^23, i.e. max value * 256
			if (ov < agc) agc = ov;    // set agc to number of 1/256ths to multiply the sample by
			z = (z1 * agc) >> 8;      // reduce sample by agc value to prevent overflow
		} else if (z <= -32768) {
			ov = -8388608/z1 - 1;
			if (ov < agc) agc = ov;
			z = (z1 * agc) >> 8;
		}
		*out_ptr++ = z;
		*out_ptr++ = z >> 8;
		if(output_hooks && output_hooks->outputVoiced) output_hooks->outputVoiced(z);

		echo_buf[echo_head++] = z;
		if (echo_head >= N_ECHO_BUF)
			echo_head = 0;

		if (out_ptr + 2 > out_end)
			return 1;
	}
}

static int PlaySilence(int length, bool resume)
{
	static int n_samples;

	nsamples = 0;
	samplecount = 0;
	wavephase = 0x7fffffff;

	if (length == 0)
		return 0;

	if (resume == false)
		n_samples = length;

	int value = 0;
	while (n_samples-- > 0) {
		value = (echo_buf[echo_tail++] * echo_amp) >> 8;

		if (echo_tail >= N_ECHO_BUF)
			echo_tail = 0;

		*out_ptr++ = value;
		*out_ptr++ = value >> 8;
		if(output_hooks && output_hooks->outputSilence) output_hooks->outputSilence(value);

		echo_buf[echo_head++] = value;
		if (echo_head >= N_ECHO_BUF)
			echo_head = 0;

		if (out_ptr + 2 > out_end)
			return 1;
	}
	return 0;
}

static int PlayWave(int length, bool resume, unsigned char *data, int scale, int amp)
{
	static int n_samples;
	static int ix = 0;
	int value;
	signed char c;

	if (resume == false) {
		n_samples = length;
		ix = 0;
	}

	nsamples = 0;
	samplecount = 0;

	while (n_samples-- > 0) {
		if (scale == 0) {
			// 16 bits data
			c = data[ix+1];
			value = data[ix] + (c * 256);
			ix += 2;
		} else {
			// 8 bit data, shift by the specified scale factor
			value = (signed char)data[ix++] * scale;
		}
		value *= (consonant_amp * general_amplitude); // reduce strength of consonant
		value = value >> 10;
		value = (value * amp)/32;

		value += ((echo_buf[echo_tail++] * echo_amp) >> 8);

		if (value > 32767)
			value = 32767;
		else if (value < -32768)
			value = -32768;

		if (echo_tail >= N_ECHO_BUF)
			echo_tail = 0;

		out_ptr[0] = value;
		out_ptr[1] = value >> 8;
		if(output_hooks && output_hooks->outputUnvoiced) output_hooks->outputUnvoiced(value);
		out_ptr += 2;

		echo_buf[echo_head++] = (value*3)/4;
		if (echo_head >= N_ECHO_BUF)
			echo_head = 0;

		if (out_ptr + 2 > out_end)
			return 1;
	}
	return 0;
}

static int SetWithRange0(int value, int max)
{
	if (value < 0)
		return 0;
	if (value > max)
		return max;
	return value;
}

static void SetPitchFormants(void)
{
	if (wvoice == NULL)
		return;

	int ix;
	int factor = 256;
	int pitch_value;

	// adjust formants to give better results for a different voice pitch
	if ((pitch_value = embedded_value[EMBED_P]) > MAX_PITCH_VALUE)
		pitch_value = MAX_PITCH_VALUE;

	if (pitch_value > 50) {
		// only adjust if the pitch is higher than normal
		factor = 256 + (25 * (pitch_value - 50))/50;
	}

	for (ix = 0; ix <= 5; ix++)
		wvoice->freq[ix] = (wvoice->freq2[ix] * factor)/256;

	factor = embedded_value[EMBED_T]*3;
	wvoice->height[0] = (wvoice->height2[0] * (256 - factor*2))/256;
	wvoice->height[1] = (wvoice->height2[1] * (256 - factor))/256;
}

void SetEmbedded(int control, int value)
{
	// there was an embedded command in the text at this point
	int sign = 0;
	int command;

	command = control & 0x1f;
	if ((control & 0x60) == 0x60)
		sign = -1;
	else if ((control & 0x60) == 0x40)
		sign = 1;

	if (command < N_EMBEDDED_VALUES) {
		if (sign == 0)
			embedded_value[command] = value;
		else
			embedded_value[command] += (value * sign);
		embedded_value[command] = SetWithRange0(embedded_value[command], embedded_max[command]);
	}

	switch (command)
	{
	case EMBED_T:
		WavegenSetEcho(); // and drop through to case P
	case EMBED_P:
		SetPitchFormants();
		break;
	case EMBED_A: // amplitude
		general_amplitude = GetAmplitude();
		break;
	case EMBED_F: // emphasis
		general_amplitude = GetAmplitude();
		break;
	case EMBED_H:
		WavegenSetEcho();
		break;
	}
}

void WavegenSetVoice(voice_t *v)
{
	static voice_t v2;

	memcpy(&v2, v, sizeof(v2));
	wvoice = &v2;

	if (v->peak_shape == 0)
		pk_shape = pk_shape1;
	else
		pk_shape = pk_shape2;

	consonant_amp = (v->consonant_amp * 26) /100;
	if (samplerate <= 11000) {
		consonant_amp = consonant_amp*2; // emphasize consonants at low sample rates
		option_harmonic1 = 6;
	}
	WavegenSetEcho();
	SetPitchFormants();
	MarkerEvent(espeakEVENT_SAMPLERATE, 0, wvoice->samplerate, 0, out_ptr);
}

static void SetAmplitude(int length, unsigned char *amp_env, int value)
{
	if (wvoice == NULL)
		return;

	amp_ix = 0;
	if (length == 0)
		amp_inc = 0;
	else
		amp_inc = (256 * ENV_LEN * STEPSIZE)/length;

	wdata.amplitude = (value * general_amplitude)/16;
	wdata.amplitude_v = (wdata.amplitude * wvoice->consonant_ampv * 15)/100; // for wave mixed with voiced sounds

	amplitude_env = amp_env;
}

void SetPitch2(voice_t *voice, int pitch1, int pitch2, int *pitch_base, int *pitch_range)
{
	int base;
	int range;
	int pitch_value;

	if (pitch1 > pitch2) {
		int x;
		x = pitch1; // swap values
		pitch1 = pitch2;
		pitch2 = x;
	}

	if ((pitch_value = embedded_value[EMBED_P]) > MAX_PITCH_VALUE)
		pitch_value = MAX_PITCH_VALUE;
	pitch_value -= embedded_value[EMBED_T]; // adjust tone for announcing punctuation
	if (pitch_value < 0)
		pitch_value = 0;

	base = (voice->pitch_base * pitch_adjust_tab[pitch_value])/128;
	range =  (voice->pitch_range * embedded_value[EMBED_R])/50;

	// compensate for change in pitch when the range is narrowed or widened
	base -= (range - voice->pitch_range)*18;

	*pitch_base = base + (pitch1 * range)/2;
	*pitch_range = base + (pitch2 * range)/2 - *pitch_base;
}

static void SetPitch(int length, unsigned char *env, int pitch1, int pitch2)
{
	if (wvoice == NULL)
		return;

	// length in samples

	if ((wdata.pitch_env = env) == NULL)
		wdata.pitch_env = env_fall; // default

	wdata.pitch_ix = 0;
	if (length == 0)
		wdata.pitch_inc = 0;
	else
		wdata.pitch_inc = (256 * ENV_LEN * STEPSIZE)/length;

	SetPitch2(wvoice, pitch1, pitch2, &wdata.pitch_base, &wdata.pitch_range);
	// set initial pitch
	wdata.pitch = ((wdata.pitch_env[0] * wdata.pitch_range) >>8) + wdata.pitch_base; // Hz << 12

	flutter_amp = wvoice->flutter;
}

static void SetSynth(int length, int modn, frame_t *fr1, frame_t *fr2, voice_t *v)
{
	if (wvoice == NULL || v == NULL)
		return;

	int ix;
	double next;
	int length2;
	int length4;
	int qix;
	static const int glottal_reduce_tab1[4] = { 0x30, 0x30, 0x40, 0x50 }; // vowel before [?], amp * 1/256
	static const int glottal_reduce_tab2[4] = { 0x90, 0xa0, 0xb0, 0xc0 }; // vowel after [?], amp * 1/256

	end_wave = 1;

	// any additional information in the param1 ?
	modulation_type = modn & 0xff;

	glottal_flag = 0;
	if (modn & 0x400) {
		glottal_flag = 3; // before a glottal stop
		glottal_reduce = glottal_reduce_tab1[(modn >> 8) & 3];
	}
	if (modn & 0x800) {
		glottal_flag = 4; // after a glottal stop
		glottal_reduce = glottal_reduce_tab2[(modn >> 8) & 3];
	}

	for (qix = wcmdq_head+1;; qix++) {
		if (qix >= N_WCMDQ) qix = 0;
		if (qix == wcmdq_tail) break;

		int cmd = wcmdq[qix][0];
		if (cmd == WCMD_SPECT) {
			end_wave = 0; // next wave generation is from another spectrum
			break;
		}
		if ((cmd == WCMD_WAVE) || (cmd == WCMD_PAUSE))
			break; // next is not from spectrum, so continue until end of wave cycle
	}

	// round the length to a multiple of the stepsize
	length2 = (length + STEPSIZE/2) & ~0x3f;
	if (length2 == 0)
		length2 = STEPSIZE;

	// add this length to any left over from the previous synth
	samplecount_start = samplecount;
	nsamples += length2;

	length4 = length2/4;

	peaks[7].freq = (7800  * v->freq[7] + v->freqadd[7]*256) << 8;
	peaks[8].freq = (9000  * v->freq[8] + v->freqadd[8]*256) << 8;

	for (ix = 0; ix < 8; ix++) {
		if (ix < 7) {
			peaks[ix].freq1 = (fr1->ffreq[ix] * v->freq[ix] + v->freqadd[ix]*256) << 8;
			peaks[ix].freq = (int)peaks[ix].freq1;
			next = (fr2->ffreq[ix] * v->freq[ix] + v->freqadd[ix]*256) << 8;
			peaks[ix].freq_inc =  ((next - peaks[ix].freq1) * (STEPSIZE/4)) / length4; // lower headroom for fixed point math
		}

		peaks[ix].height1 = (fr1->fheight[ix] * v->height[ix]) << 6;
		peaks[ix].height = (int)peaks[ix].height1;
		next = (fr2->fheight[ix] * v->height[ix]) << 6;
		peaks[ix].height_inc =  ((next - peaks[ix].height1) * STEPSIZE) / length2;

		if ((ix <= 5) && (ix <= wvoice->n_harmonic_peaks)) {
			peaks[ix].left1 = (fr1->fwidth[ix] * v->width[ix]) << 10;
			peaks[ix].left = (int)peaks[ix].left1;
			next = (fr2->fwidth[ix] * v->width[ix]) << 10;
			peaks[ix].left_inc =  ((next - peaks[ix].left1) * STEPSIZE) / length2;

			if (ix < 3) {
				peaks[ix].right1 = (fr1->fright[ix] * v->width[ix]) << 10;
				peaks[ix].right = (int)peaks[ix].right1;
				next = (fr2->fright[ix] * v->width[ix]) << 10;
				peaks[ix].right_inc = ((next - peaks[ix].right1) * STEPSIZE) / length2;
			} else
				peaks[ix].right = peaks[ix].left;
		}
	}
}

void Write4Bytes(FILE *f, int value)
{
	// Write 4 bytes to a file, least significant first
	int ix;

	for (ix = 0; ix < 4; ix++) {
		fputc(value & 0xff, f);
		value = value >> 8;
	}
}

static int WavegenFill2(void)
{
	// Pick up next wavegen commands from the queue
	// return: 0  output buffer has been filled
	// return: 1  input command queue is now empty
	intptr_t *q;
	int length;
	int result;
	int marker_type;
	static bool resume = false;
	static int echo_complete = 0;

	while (out_ptr < out_end) {
		if (WcmdqUsed() <= 0) {
			if (echo_complete > 0) {
				// continue to play silence until echo is completed
				resume = PlaySilence(echo_complete, resume);
				if (resume == true)
					return 0; // not yet finished
			}
			return 1; // queue empty, close sound channel
		}

		result = 0;
		q = wcmdq[wcmdq_head];
		length = q[1];

		switch (q[0] & 0xff)
		{
		case WCMD_PITCH:
			SetPitch(length, (unsigned char *)q[2], q[3] >> 16, q[3] & 0xffff);
			break;
		case WCMD_PHONEME_ALIGNMENT:
		{
			char* data = (char*)q[1];
			output_hooks->outputPhoSymbol(data,q[2]);
			free(data);
		}
			break;
		case WCMD_PAUSE:
			if (resume == false)
				echo_complete -= length;
			wdata.n_mix_wavefile = 0;
			wdata.amplitude_fmt = 100;
#if USE_KLATT
			KlattReset(1);
#endif
			result = PlaySilence(length, resume);
			break;
		case WCMD_WAVE:
			echo_complete = echo_length;
			wdata.n_mix_wavefile = 0;
#if USE_KLATT
			KlattReset(1);
#endif
			result = PlayWave(length, resume, (unsigned char *)q[2], q[3] & 0xff, q[3] >> 8);
			break;
		case WCMD_WAVE2:
			// wave file to be played at the same time as synthesis
			wdata.mix_wave_amp = q[3] >> 8;
			wdata.mix_wave_scale = q[3] & 0xff;
			wdata.n_mix_wavefile = (length & 0xffff);
			wdata.mix_wavefile_max = (length >> 16) & 0xffff;
			if (wdata.mix_wave_scale == 0) {
				wdata.n_mix_wavefile *= 2;
				wdata.mix_wavefile_max *= 2;
			}
			wdata.mix_wavefile_ix = 0;
			wdata.mix_wavefile_offset = 0;
			wdata.mix_wavefile = (unsigned char *)q[2];
			break;
		case WCMD_SPECT2: // as WCMD_SPECT but stop any concurrent wave file
			wdata.n_mix_wavefile = 0; // ... and drop through to WCMD_SPECT case
		case WCMD_SPECT:
			echo_complete = echo_length;
			result = Wavegen(length & 0xffff, q[1] >> 16, resume, (frame_t *)q[2], (frame_t *)q[3], wvoice);
			break;
#if USE_KLATT
		case WCMD_KLATT2: // as WCMD_SPECT but stop any concurrent wave file
			wdata.n_mix_wavefile = 0; // ... and drop through to WCMD_SPECT case
		case WCMD_KLATT:
			echo_complete = echo_length;
			result = Wavegen_Klatt(length & 0xffff, resume, (frame_t *)q[2], (frame_t *)q[3], &wdata, wvoice);
			break;
#endif
		case WCMD_MARKER:
			marker_type = q[0] >> 8;
			MarkerEvent(marker_type, q[1], q[2], q[3], out_ptr);
			break;
		case WCMD_AMPLITUDE:
			SetAmplitude(length, (unsigned char *)q[2], q[3]);
			break;
		case WCMD_VOICE:
			WavegenSetVoice((voice_t *)q[2]);
			free((voice_t *)q[2]);
			break;
		case WCMD_EMBEDDED:
			SetEmbedded(q[1], q[2]);
			break;
#if USE_MBROLA
		case WCMD_MBROLA_DATA:
			if (wvoice != NULL)
				result = MbrolaFill(length, resume, (general_amplitude * wvoice->voicing)/64);
			break;
#endif
		case WCMD_FMT_AMPLITUDE:
			if ((wdata.amplitude_fmt = q[1]) == 0)
				wdata.amplitude_fmt = 100; // percentage, but value=0 means 100%
			break;
#if USE_LIBSONIC
		case WCMD_SONIC_SPEED:
			sonicSpeed = (double)q[1] / 1024;
			if (sonicSpeedupStream && (sonicSpeed <= 1.0)) {
				sonicFlushStream(sonicSpeedupStream);
				int length = (out_end - out_ptr);
				length = sonicReadShortFromStream(sonicSpeedupStream, (short*)out_ptr, length/2);
				out_ptr += length * 2;
			}
			break;
#endif
		}

		if (result == 0) {
			WcmdqIncHead();
			resume = false;
		} else
			resume = true;
	}

	return 0;
}

#if USE_LIBSONIC
// Speed up the audio samples with libsonic.
static int SpeedUp(short *outbuf, int length_in, int length_out, int end_of_text)
{
	if (length_in > 0) {
		if (sonicSpeedupStream == NULL)
			sonicSpeedupStream = sonicCreateStream(22050, 1);
		if (sonicGetSpeed(sonicSpeedupStream) != sonicSpeed)
			sonicSetSpeed(sonicSpeedupStream, sonicSpeed);

		sonicWriteShortToStream(sonicSpeedupStream, outbuf, length_in);
	}

	if (sonicSpeedupStream == NULL)
		return 0;

	if (end_of_text)
		sonicFlushStream(sonicSpeedupStream);
	return sonicReadShortFromStream(sonicSpeedupStream, outbuf, length_out);
}
#endif

// Call WavegenFill2, and then speed up the output samples.
int WavegenFill(void)
{
	int finished;
#if USE_LIBSONIC
	unsigned char *p_start;

	p_start = out_ptr;
#endif

	finished = WavegenFill2();

#if USE_LIBSONIC
	if (sonicSpeed > 1.0) {
		int length;
		int max_length;

		max_length = (out_end - p_start);
		length =  2*SpeedUp((short *)p_start, (out_ptr-p_start)/2, max_length/2, finished);
		out_ptr = p_start + length;

		if (length >= max_length)
			finished = 0; // there may be more data to flush
	}
#endif
	return finished;
}

#pragma GCC visibility push(default)

ESPEAK_NG_API espeak_ng_STATUS
espeak_ng_SetOutputHooks(espeak_ng_OUTPUT_HOOKS* hooks)
{
	output_hooks = hooks;
	return 0;
}

ESPEAK_NG_API espeak_ng_STATUS
espeak_ng_SetConstF0(int f0)
{
	const_f0 = f0;
	return ENS_OK;
}

#pragma GCC visibility pop