You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
687 lines
16 KiB
687 lines
16 KiB
/* |
|
s_vox.c - npc sentences |
|
Copyright (C) 2010 Uncle Mike |
|
|
|
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. |
|
*/ |
|
|
|
#include "common.h" |
|
#include "sound.h" |
|
#include "const.h" |
|
|
|
sentence_t g_Sentences[MAX_SENTENCES]; |
|
static uint g_numSentences; |
|
static char *rgpparseword[CVOXWORDMAX]; // array of pointers to parsed words |
|
static char voxperiod[] = "_period"; // vocal pause |
|
static char voxcomma[] = "_comma"; // vocal pause |
|
|
|
static int IsNextWord( const char c ) |
|
{ |
|
if( c == '.' || c == ',' || c == ' ' || c == '(' ) |
|
return 1; |
|
return 0; |
|
} |
|
|
|
static int IsSkipSpace( const char c ) |
|
{ |
|
if( c == ',' || c == '.' || c == ' ' ) |
|
return 1; |
|
return 0; |
|
} |
|
|
|
static int IsWhiteSpace( const char space ) |
|
{ |
|
if( space == ' ' || space == '\t' || space == '\r' || space == '\n' ) |
|
return 1; |
|
return 0; |
|
} |
|
|
|
static int IsCommandChar( const char c ) |
|
{ |
|
if( c == 'v' || c == 'p' || c == 's' || c == 'e' || c == 't' ) |
|
return 1; |
|
return 0; |
|
} |
|
|
|
static int IsDelimitChar( const char c ) |
|
{ |
|
if( c == '(' || c == ')' ) |
|
return 1; |
|
return 0; |
|
} |
|
|
|
static char *ScanForwardUntil( char *string, const char scan ) |
|
{ |
|
while( string[0] ) |
|
{ |
|
if( string[0] == scan ) |
|
return string; |
|
string++; |
|
} |
|
return string; |
|
} |
|
|
|
// backwards scan psz for last '/' |
|
// return substring in szpath null terminated |
|
// if '/' not found, return 'vox/' |
|
static char *VOX_GetDirectory( char *szpath, char *psz ) |
|
{ |
|
char c; |
|
int cb = 0; |
|
char *p = psz + Q_strlen( psz ) - 1; |
|
|
|
// scan backwards until first '/' or start of string |
|
c = *p; |
|
while( p > psz && c != '/' ) |
|
{ |
|
c = *( --p ); |
|
cb++; |
|
} |
|
|
|
if( c != '/' ) |
|
{ |
|
// didn't find '/', return default directory |
|
Q_strcpy( szpath, "vox/" ); |
|
return psz; |
|
} |
|
|
|
cb = Q_strlen( psz ) - cb; |
|
memcpy( szpath, psz, cb ); |
|
szpath[cb] = 0; |
|
|
|
return p + 1; |
|
} |
|
|
|
// scan g_Sentences, looking for pszin sentence name |
|
// return pointer to sentence data if found, null if not |
|
// CONSIDER: if we have a large number of sentences, should |
|
// CONSIDER: sort strings in g_Sentences and do binary search. |
|
char *VOX_LookupString( const char *pSentenceName, int *psentencenum ) |
|
{ |
|
int i; |
|
|
|
if( Q_isdigit( pSentenceName ) && (i = Q_atoi( pSentenceName )) < g_numSentences ) |
|
{ |
|
if( psentencenum ) *psentencenum = i; |
|
return (g_Sentences[i].pName + Q_strlen( g_Sentences[i].pName ) + 1 ); |
|
} |
|
|
|
for( i = 0; i < g_numSentences; i++ ) |
|
{ |
|
if( !Q_stricmp( pSentenceName, g_Sentences[i].pName )) |
|
{ |
|
if( psentencenum ) *psentencenum = i; |
|
return (g_Sentences[i].pName + Q_strlen( g_Sentences[i].pName ) + 1 ); |
|
} |
|
} |
|
|
|
return NULL; |
|
} |
|
|
|
// parse a null terminated string of text into component words, with |
|
// pointers to each word stored in rgpparseword |
|
// note: this code actually alters the passed in string! |
|
char **VOX_ParseString( char *psz ) |
|
{ |
|
int i, fdone = 0; |
|
char c, *p = psz; |
|
|
|
memset( rgpparseword, 0, sizeof( char* ) * CVOXWORDMAX ); |
|
|
|
if( !psz ) return NULL; |
|
|
|
i = 0; |
|
rgpparseword[i++] = psz; |
|
|
|
while( !fdone && i < CVOXWORDMAX ) |
|
{ |
|
// scan up to next word |
|
c = *p; |
|
while( c && !IsNextWord( c )) |
|
c = *(++p); |
|
|
|
// if '(' then scan for matching ')' |
|
if( c == '(' ) |
|
{ |
|
p = ScanForwardUntil( p, ')' ); |
|
c = *(++p); |
|
if( !c ) fdone = 1; |
|
} |
|
|
|
if( fdone || !c ) |
|
{ |
|
fdone = 1; |
|
} |
|
else |
|
{ |
|
// if . or , insert pause into rgpparseword, |
|
// unless this is the last character |
|
if(( c == '.' || c == ',' ) && *(p+1) != '\n' && *(p+1) != '\r' && *(p+1) != 0 ) |
|
{ |
|
if( c == '.' ) rgpparseword[i++] = voxperiod; |
|
else rgpparseword[i++] = voxcomma; |
|
|
|
if( i >= CVOXWORDMAX ) |
|
break; |
|
} |
|
|
|
// null terminate substring |
|
*p++ = 0; |
|
|
|
// skip whitespace |
|
c = *p; |
|
while( c && IsSkipSpace( c )) |
|
c = *(++p); |
|
|
|
if( !c ) fdone = 1; |
|
else rgpparseword[i++] = p; |
|
} |
|
} |
|
|
|
return rgpparseword; |
|
} |
|
|
|
float VOX_GetVolumeScale( channel_t *pchan ) |
|
{ |
|
if( pchan->currentWord ) |
|
{ |
|
if ( pchan->words[pchan->wordIndex].volume ) |
|
{ |
|
float volume = pchan->words[pchan->wordIndex].volume * 0.01f; |
|
if( volume < 1.0f ) return volume; |
|
} |
|
} |
|
|
|
return 1.0f; |
|
} |
|
|
|
void VOX_SetChanVol( channel_t *ch ) |
|
{ |
|
float scale; |
|
|
|
if( !ch->currentWord ) |
|
return; |
|
|
|
scale = VOX_GetVolumeScale( ch ); |
|
if( scale == 1.0f ) return; |
|
|
|
ch->rightvol = (int)(ch->rightvol * scale); |
|
ch->leftvol = (int)(ch->leftvol * scale); |
|
} |
|
|
|
float VOX_ModifyPitch( channel_t *ch, float pitch ) |
|
{ |
|
if( ch->currentWord ) |
|
{ |
|
if( ch->words[ch->wordIndex].pitch > 0 ) |
|
{ |
|
pitch += ( ch->words[ch->wordIndex].pitch - PITCH_NORM ) * 0.01f; |
|
} |
|
} |
|
|
|
return pitch; |
|
} |
|
|
|
//=============================================================================== |
|
// Get any pitch, volume, start, end params into voxword |
|
// and null out trailing format characters |
|
// Format: |
|
// someword(v100 p110 s10 e20) |
|
// |
|
// v is volume, 0% to n% |
|
// p is pitch shift up 0% to n% |
|
// s is start wave offset % |
|
// e is end wave offset % |
|
// t is timecompression % |
|
// |
|
// pass fFirst == 1 if this is the first string in sentence |
|
// returns 1 if valid string, 0 if parameter block only. |
|
// |
|
// If a ( xxx ) parameter block does not directly follow a word, |
|
// then that 'default' parameter block will be used as the default value |
|
// for all following words. Default parameter values are reset |
|
// by another 'default' parameter block. Default parameter values |
|
// for a single word are overridden for that word if it has a parameter block. |
|
// |
|
//=============================================================================== |
|
int VOX_ParseWordParams( char *psz, voxword_t *pvoxword, int fFirst ) |
|
{ |
|
char *pszsave = psz; |
|
char c, ct, sznum[8]; |
|
static voxword_t voxwordDefault; |
|
int i; |
|
|
|
// init to defaults if this is the first word in string. |
|
if( fFirst ) |
|
{ |
|
voxwordDefault.pitch = -1; |
|
voxwordDefault.volume = 100; |
|
voxwordDefault.start = 0; |
|
voxwordDefault.end = 100; |
|
voxwordDefault.fKeepCached = 0; |
|
voxwordDefault.timecompress = 0; |
|
} |
|
|
|
*pvoxword = voxwordDefault; |
|
|
|
// look at next to last char to see if we have a |
|
// valid format: |
|
c = *( psz + Q_strlen( psz ) - 1 ); |
|
|
|
// no formatting, return |
|
if( c != ')' ) return 1; |
|
|
|
// scan forward to first '(' |
|
c = *psz; |
|
while( !IsDelimitChar( c )) |
|
c = *(++psz); |
|
|
|
// bogus formatting |
|
if( c == ')' ) return 0; |
|
|
|
// null terminate |
|
*psz = 0; |
|
ct = *(++psz); |
|
|
|
while( 1 ) |
|
{ |
|
// scan until we hit a character in the commandSet |
|
while( ct && !IsCommandChar( ct )) |
|
ct = *(++psz); |
|
|
|
if( ct == ')' ) |
|
break; |
|
|
|
memset( sznum, 0, sizeof( sznum )); |
|
i = 0; |
|
|
|
c = *(++psz); |
|
|
|
if( !isdigit( c )) |
|
break; |
|
|
|
// read number |
|
while( isdigit( c ) && i < sizeof( sznum ) - 1 ) |
|
{ |
|
sznum[i++] = c; |
|
c = *(++psz); |
|
} |
|
|
|
// get value of number |
|
i = Q_atoi( sznum ); |
|
|
|
switch( ct ) |
|
{ |
|
case 'v': pvoxword->volume = i; break; |
|
case 'p': pvoxword->pitch = i; break; |
|
case 's': pvoxword->start = i; break; |
|
case 'e': pvoxword->end = i; break; |
|
case 't': pvoxword->timecompress = i; break; |
|
} |
|
|
|
ct = c; |
|
} |
|
|
|
// if the string has zero length, this was an isolated |
|
// parameter block. Set default voxword to these |
|
// values |
|
if( Q_strlen( pszsave ) == 0 ) |
|
{ |
|
voxwordDefault = *pvoxword; |
|
return 0; |
|
} |
|
|
|
return 1; |
|
} |
|
|
|
void VOX_LoadWord( channel_t *pchan ) |
|
{ |
|
if( pchan->words[pchan->wordIndex].sfx ) |
|
{ |
|
wavdata_t *pSource = S_LoadSound( pchan->words[pchan->wordIndex].sfx ); |
|
|
|
if( pSource ) |
|
{ |
|
int start = pchan->words[pchan->wordIndex].start; |
|
int end = pchan->words[pchan->wordIndex].end; |
|
|
|
// apply mixer |
|
pchan->currentWord = &pchan->pMixer; |
|
pchan->currentWord->pData = pSource; |
|
|
|
// don't allow overlapped ranges |
|
if( end <= start ) end = 0; |
|
|
|
if( start || end ) |
|
{ |
|
int sampleCount = pSource->samples; |
|
|
|
if( start ) |
|
{ |
|
S_SetSampleStart( pchan, pSource, (int)(sampleCount * 0.01f * start)); |
|
} |
|
|
|
if( end ) |
|
{ |
|
S_SetSampleEnd( pchan, pSource, (int)(sampleCount * 0.01f * end)); |
|
} |
|
} |
|
} |
|
} |
|
} |
|
|
|
void VOX_FreeWord( channel_t *pchan ) |
|
{ |
|
pchan->currentWord = NULL; // sentence is finished |
|
memset( &pchan->pMixer, 0, sizeof( pchan->pMixer )); |
|
|
|
// release unused sounds |
|
if( pchan->words[pchan->wordIndex].sfx ) |
|
{ |
|
// If this wave wasn't precached by the game code |
|
if( !pchan->words[pchan->wordIndex].fKeepCached ) |
|
{ |
|
FS_FreeSound( pchan->words[pchan->wordIndex].sfx->cache ); |
|
pchan->words[pchan->wordIndex].sfx->cache = NULL; |
|
pchan->words[pchan->wordIndex].sfx = NULL; |
|
} |
|
} |
|
} |
|
|
|
void VOX_LoadFirstWord( channel_t *pchan, voxword_t *pwords ) |
|
{ |
|
int i = 0; |
|
|
|
// copy each pointer in the sfx temp array into the |
|
// sentence array, and set the channel to point to the |
|
// sentence array |
|
while( pwords[i].sfx != NULL ) |
|
{ |
|
pchan->words[i] = pwords[i]; |
|
i++; |
|
} |
|
pchan->words[i].sfx = NULL; |
|
|
|
pchan->wordIndex = 0; |
|
VOX_LoadWord( pchan ); |
|
} |
|
|
|
// return number of samples mixed |
|
int VOX_MixDataToDevice( channel_t *pchan, int sampleCount, int outputRate, int outputOffset ) |
|
{ |
|
// save this to compute total output |
|
int startingOffset = outputOffset; |
|
|
|
if( !pchan->currentWord ) |
|
return 0; |
|
|
|
while( sampleCount > 0 && pchan->currentWord ) |
|
{ |
|
int timeCompress = pchan->words[pchan->wordIndex].timecompress; |
|
int outputCount = S_MixDataToDevice( pchan, sampleCount, outputRate, outputOffset, timeCompress ); |
|
|
|
outputOffset += outputCount; |
|
sampleCount -= outputCount; |
|
|
|
// if we finished load a next word |
|
if( pchan->currentWord->finished ) |
|
{ |
|
VOX_FreeWord( pchan ); |
|
pchan->wordIndex++; |
|
VOX_LoadWord( pchan ); |
|
|
|
if( pchan->currentWord ) |
|
{ |
|
pchan->sfx = pchan->words[pchan->wordIndex].sfx; |
|
} |
|
} |
|
} |
|
return outputOffset - startingOffset; |
|
} |
|
|
|
// link all sounds in sentence, start playing first word. |
|
void VOX_LoadSound( channel_t *pchan, const char *pszin ) |
|
{ |
|
char buffer[512]; |
|
int i, cword; |
|
char pathbuffer[64]; |
|
char szpath[32]; |
|
voxword_t rgvoxword[CVOXWORDMAX]; |
|
char *psz; |
|
|
|
if( !pszin || !*pszin ) |
|
return; |
|
|
|
memset( rgvoxword, 0, sizeof( voxword_t ) * CVOXWORDMAX ); |
|
memset( buffer, 0, sizeof( buffer )); |
|
|
|
// lookup actual string in g_Sentences, |
|
// set pointer to string data |
|
psz = VOX_LookupString( pszin, NULL ); |
|
|
|
if( !psz ) |
|
{ |
|
Con_DPrintf( S_ERROR "VOX_LoadSound: no such sentence %s\n", pszin ); |
|
return; |
|
} |
|
|
|
// get directory from string, advance psz |
|
psz = VOX_GetDirectory( szpath, psz ); |
|
|
|
if( Q_strlen( psz ) > sizeof( buffer ) - 1 ) |
|
{ |
|
Con_Printf( S_ERROR "VOX_LoadSound: sentence is too long %s\n", psz ); |
|
return; |
|
} |
|
|
|
// copy into buffer |
|
Q_strcpy( buffer, psz ); |
|
psz = buffer; |
|
|
|
// parse sentence (also inserts null terminators between words) |
|
VOX_ParseString( psz ); |
|
|
|
// for each word in the sentence, construct the filename, |
|
// lookup the sfx and save each pointer in a temp array |
|
|
|
i = 0; |
|
cword = 0; |
|
while( rgpparseword[i] ) |
|
{ |
|
// Get any pitch, volume, start, end params into voxword |
|
if( VOX_ParseWordParams( rgpparseword[i], &rgvoxword[cword], i == 0 )) |
|
{ |
|
// this is a valid word (as opposed to a parameter block) |
|
Q_strcpy( pathbuffer, szpath ); |
|
Q_strncat( pathbuffer, rgpparseword[i], sizeof( pathbuffer )); |
|
Q_strncat( pathbuffer, ".wav", sizeof( pathbuffer )); |
|
|
|
// find name, if already in cache, mark voxword |
|
// so we don't discard when word is done playing |
|
rgvoxword[cword].sfx = S_FindName( pathbuffer, &( rgvoxword[cword].fKeepCached )); |
|
cword++; |
|
} |
|
i++; |
|
} |
|
|
|
VOX_LoadFirstWord( pchan, rgvoxword ); |
|
|
|
pchan->isSentence = true; |
|
pchan->sfx = rgvoxword[0].sfx; |
|
} |
|
|
|
//----------------------------------------------------------------------------- |
|
// Purpose: Take a NULL terminated sentence, and parse any commands contained in |
|
// {}. The string is rewritten in place with those commands removed. |
|
// |
|
// Input : *pSentenceData - sentence data to be modified in place |
|
// sentenceIndex - global sentence table index for any data that is |
|
// parsed out |
|
//----------------------------------------------------------------------------- |
|
void VOX_ParseLineCommands( char *pSentenceData, int sentenceIndex ) |
|
{ |
|
char tempBuffer[512]; |
|
char *pNext, *pStart; |
|
int length, tempBufferPos = 0; |
|
|
|
if( !pSentenceData ) |
|
return; |
|
|
|
pStart = pSentenceData; |
|
|
|
while( *pSentenceData ) |
|
{ |
|
pNext = ScanForwardUntil( pSentenceData, '{' ); |
|
|
|
// find length of "good" portion of the string (not a {} command) |
|
length = pNext - pSentenceData; |
|
if( tempBufferPos + length > sizeof( tempBuffer )) |
|
{ |
|
Con_Printf( S_ERROR "Sentence too long (max length %d characters)\n", sizeof(tempBuffer) - 1 ); |
|
return; |
|
} |
|
|
|
// Copy good string to temp buffer |
|
memcpy( tempBuffer + tempBufferPos, pSentenceData, length ); |
|
|
|
// move the copy position |
|
tempBufferPos += length; |
|
|
|
pSentenceData = pNext; |
|
|
|
// skip ahead of the opening brace |
|
if( *pSentenceData ) pSentenceData++; |
|
|
|
// skip whitespace |
|
while( *pSentenceData && *pSentenceData <= 32 ) |
|
pSentenceData++; |
|
|
|
// simple comparison of string commands: |
|
switch( Q_tolower( *pSentenceData )) |
|
{ |
|
case 'l': |
|
// all commands starting with the letter 'l' here |
|
if( !Q_strnicmp( pSentenceData, "len", 3 )) |
|
{ |
|
g_Sentences[sentenceIndex].length = Q_atof( pSentenceData + 3 ); |
|
} |
|
break; |
|
case 0: |
|
default: |
|
break; |
|
} |
|
|
|
pSentenceData = ScanForwardUntil( pSentenceData, '}' ); |
|
|
|
// skip the closing brace |
|
if( *pSentenceData ) pSentenceData++; |
|
|
|
// skip trailing whitespace |
|
while( *pSentenceData && *pSentenceData <= 32 ) |
|
pSentenceData++; |
|
} |
|
|
|
if( tempBufferPos < sizeof( tempBuffer )) |
|
{ |
|
// terminate cleaned up copy |
|
tempBuffer[tempBufferPos] = 0; |
|
|
|
// copy it over the original data |
|
Q_strcpy( pStart, tempBuffer ); |
|
} |
|
} |
|
|
|
// Load sentence file into memory, insert null terminators to |
|
// delimit sentence name/sentence pairs. Keep pointer to each |
|
// sentence name so we can search later. |
|
void VOX_ReadSentenceFile( const char *psentenceFileName ) |
|
{ |
|
char c, *pch, *pFileData; |
|
char *pchlast, *pSentenceData; |
|
fs_offset_t fileSize; |
|
|
|
// load file |
|
pFileData = (char *)FS_LoadFile( psentenceFileName, &fileSize, false ); |
|
if( !pFileData ) return; // this game just doesn't used vox sound system |
|
|
|
pch = pFileData; |
|
pchlast = pch + fileSize; |
|
|
|
while( pch < pchlast ) |
|
{ |
|
if( g_numSentences >= MAX_SENTENCES ) |
|
{ |
|
Con_Printf( S_ERROR "VOX_Init: too many sentences specified, max is %d\n", MAX_SENTENCES ); |
|
break; |
|
} |
|
|
|
// only process this pass on sentences |
|
pSentenceData = NULL; |
|
|
|
// skip newline, cr, tab, space |
|
|
|
c = *pch; |
|
while( pch < pchlast && IsWhiteSpace( c )) |
|
c = *(++pch); |
|
|
|
// skip entire line if first char is / |
|
if( *pch != '/' ) |
|
{ |
|
sentence_t *pSentence = &g_Sentences[g_numSentences++]; |
|
|
|
pSentence->pName = pch; |
|
pSentence->length = 0; |
|
|
|
// scan forward to first space, insert null terminator |
|
// after sentence name |
|
|
|
c = *pch; |
|
while( pch < pchlast && c != ' ' ) |
|
c = *(++pch); |
|
|
|
if( pch < pchlast ) |
|
*pch++ = 0; |
|
|
|
// a sentence may have some line commands, make an extra pass |
|
pSentenceData = pch; |
|
} |
|
|
|
// scan forward to end of sentence or eof |
|
while( pch < pchlast && pch[0] != '\n' && pch[0] != '\r' ) |
|
pch++; |
|
|
|
// insert null terminator |
|
if( pch < pchlast ) *pch++ = 0; |
|
|
|
// If we have some sentence data, parse out any line commands |
|
if( pSentenceData && pSentenceData < pchlast ) |
|
{ |
|
int index = g_numSentences - 1; |
|
|
|
// the current sentence has an index of count-1 |
|
VOX_ParseLineCommands( pSentenceData, index ); |
|
} |
|
} |
|
} |
|
|
|
void VOX_Init( void ) |
|
{ |
|
memset( g_Sentences, 0, sizeof( g_Sentences )); |
|
g_numSentences = 0; |
|
|
|
VOX_ReadSentenceFile( DEFAULT_SOUNDPATH "sentences.txt" ); |
|
} |
|
|
|
|
|
void VOX_Shutdown( void ) |
|
{ |
|
g_numSentences = 0; |
|
}
|
|
|