src/bz/audio/speech.cpp

   1 #include <bz/audio/speech.h>
   2
   3 #include <bz/memory/allocator.h>
   4 #include <bz/resources/resource.h>
   5 #include <bz/types/identifier_internal.h>
   6 #include <darray.h>
   7 #include <klatt.h>
   8 #include <tts.h>
   9 #include <new>
  10 #include <parson.h>
  11 #include <string.h> // strcmp
  12
  13 struct BZAudioVoice {
  14         darray element;
  15         klatt synth;
  16         size_t maxFrames;
  17 //      short *sample;
  18 };
  19
  20 BZAudioVoice *bzAudioLoadSpeechVoice(BZAudioVoiceDetails *detailsOut, BZMemoryArenaID arena, const char *identifierFmt, ...) {
  21         bzMakeIdentifier(identifier, identifierFmt);
  22
  23         int baseFrequency = 1330;
  24         float baseSpeed = 10.0f;
  25         float baseDeclination = 0.5f;
  26         int aBaseWaveform = KW_SAW;
  27
  28         BZResourceID handle = bzResourcesOpenResource("voices", "assets/voices/%s.voice.json", identifier);
  29         size_t length = bzResourcesFileLength(handle);
  30
  31         char *data = (char *)alloca(length); // FIXME, temporary memory
  32         bzResourcesReadBytes(handle, data, length);
  33         bzResourcesCloseResource(handle);
  34
  35         //json_set_allocation_functions
  36         JSON_Value *voiceJson = json_parse_string(data);
  37         JSON_Object *voiceJsonObject = json_object(voiceJson);
  38
  39         if (json_object_has_value_of_type(voiceJsonObject, "frequency", JSONNumber)) {
  40                 baseFrequency = json_object_get_number(voiceJsonObject, "frequency");
  41         }
  42
  43         if (json_object_has_value_of_type(voiceJsonObject, "speed", JSONNumber)) {
  44                 baseSpeed = json_object_get_number(voiceJsonObject, "speed");
  45         }
  46
  47         if (json_object_has_value_of_type(voiceJsonObject, "declination", JSONNumber)) {
  48                 baseDeclination = json_object_get_number(voiceJsonObject, "declination");
  49         }
  50
  51         if (json_object_has_value_of_type(voiceJsonObject, "waveform", JSONString)) {
  52                 const char *waveform = json_object_get_string(voiceJsonObject, "waveform");
  53                 if (strcmp(waveform, "saw") == 0) {
  54                         aBaseWaveform = KW_SAW;
  55                 } else if (strcmp(waveform, "triangle") == 0) {
  56                         aBaseWaveform = KW_TRIANGLE;
  57                 } else if (strcmp(waveform, "sin") == 0) {
  58                         aBaseWaveform = KW_SIN;
  59                 } else if (strcmp(waveform, "square") == 0) {
  60                         aBaseWaveform = KW_SQUARE;
  61                 } else if (strcmp(waveform, "pulse") == 0) {
  62                         aBaseWaveform = KW_PULSE;
  63                 } else if (strcmp(waveform, "noise") == 0) {
  64                         aBaseWaveform = KW_NOISE;
  65                 } else if (strcmp(waveform, "warble") == 0) {
  66                         aBaseWaveform = KW_WARBLE;
  67                 }
  68         }
  69
  70         json_value_free(voiceJson);
  71
  72         BZAudioVoice *voice = new(bzMemoryAlloc(arena, sizeof(BZAudioVoice))) BZAudioVoice();
  73         voice->synth.init(baseFrequency, baseSpeed, baseDeclination, aBaseWaveform);
  74         voice->maxFrames = 500;//4 * 1024;
  75 //      voice->sample = (short *)bzMemoryAlloc(arena, voice->synth.mNspFr * voice->maxFrames * sizeof(short));
  76
  77         detailsOut->channels = 1;
  78         detailsOut->rate = 11025;
  79         detailsOut->maxSamples = voice->synth.mNspFr * voice->maxFrames;
  80
  81 //      *sizeOut = outputCount * sizeof(short);
  82 //      *samplingRateOut = 11025;
  83 //      *channelsOut = 1;
  84 //      *bitsPerSampleOut = 16;
  85
  86         return voice;
  87 }
  88
  89 size_t bzAudioGenerateSpeech(BZAudioVoice *voice, short *dst, size_t dstSize, const char *speechFmt, ...) {
  90         bzMakeIdentifier(speech, speechFmt);
  91
  92         voice->element.clear();
  93
  94         darray phone; // FIXME, darray should have a stable memory footprint
  95         xlate_string(speech, &phone);
  96         int frames = klatt::phone_to_elm(phone.getData(), phone.getSize(), &voice->element);
  97         bzAssertMessage(frames <= voice->maxFrames, "Too many frames, %d", frames);
  98
  99         voice->synth.initsynth(voice->element.getSize(), (unsigned char *)voice->element.getData());
 100
 101         size_t sampleSize = voice->synth.mNspFr * frames;
 102         size_t outputCount = 0;
 103         while (outputCount < sampleSize) {
 104                 outputCount += voice->synth.synth(voice->synth.mNspFr /* This seems to be ignored... */, &dst[outputCount]);
 105                 if (outputCount == sampleSize) { break; }
 106         }
 107
 108         return outputCount * sizeof(short);
 109 }