[Scummvm-tracker] [ScummVM :: Bugs] #16393: PS3: Broken Sword 1 - Distorted noise instead of speech

[ScummVM :: Bugs]

#16393: PS3: Broken Sword 1 - Distorted noise instead of speech
------------------------+-----------------------------
Reporter:  TwistedZeon  |       Owner:  (none)
    Type:  defect       |      Status:  new
Priority:  normal       |   Component:  Audio
 Version:               |  Resolution:
Keywords:               |        Game:  Broken Sword 1
------------------------+-----------------------------
Comment (by dwatteau):

 Thanks for reporting this.

 I can reproduce the issue on that OSXPPC Linux desktop I have (similar
 big-endian PowerPC processor).

 I'm using the "original version" that GOG ships in its bonus; speech &
 audio and in Ogg Vorbis, and I think the handling of this file format is
 part of this bug (see below).

 With the current code on `branch-3-0`, on my big-endian setup I got this:

 * Windows GOG release (Vorbis music & spech): FAIL
 * PSX demo: FAIL
 * Macintosh demo: speech OK, serveSample() FAIL
 * Windows demo: OK

 (the demos being hosted on our demos page.)

 If I do something like this (which is similar to what Andy committed and
 then reverted, but with a MKTAG() paranoid change, and a speech fix as
 well):

 {{{
 diff --git a/engines/sword1/sound.cpp b/engines/sword1/sound.cpp
 index d38ce46512c..eefad722e86 100644
 --- a/engines/sword1/sound.cpp
 +++ b/engines/sword1/sound.cpp
 @@ -41,7 +41,11 @@

  namespace Sword1 {

 -#define SPEECH_FLAGS (Audio::FLAG_16BITS | Audio::FLAG_LITTLE_ENDIAN)
 +#ifdef SCUMM_LITTLE_ENDIAN
 +#define AUDIO_FLAGS (Audio::FLAG_16BITS | Audio::FLAG_LITTLE_ENDIAN)
 +#else
 +#define AUDIO_FLAGS (Audio::FLAG_16BITS | Audio::FLAG_BIG_ENDIAN)
 +#endif

  Sound::Sound(Audio::Mixer *mixer, SwordEngine *vm, ResMan *pResMan)
         : _rnd("sword1sound") {
 @@ -600,7 +604,7 @@ bool Sound::expandSpeech(byte *src, byte *dst, uint32
 dstSize, bool *endiannessC
         uint8 *fBuf = src;
         uint32 headerPos = 0;

 -       while ((READ_BE_UINT32(fBuf + headerPos) != 'data') && (headerPos
 < 100))
 +       while ((READ_BE_UINT32(fBuf + headerPos) !=
 MKTAG('d','a','t','a')) && (headerPos < 100))
                 headerPos++;

         if (headerPos < 100) {
 @@ -849,7 +853,7 @@ void Sound::playSpeech() {
         _speechSampleBusy = true;

         Audio::SeekableAudioStream *stream = Audio::makeRawStream(
 -               (byte *)_speechSample, _speechSize, 11025, SPEECH_FLAGS,
 DisposeAfterUse::NO);
 +               (byte *)_speechSample, _speechSize, 11025, AUDIO_FLAGS,
 DisposeAfterUse::NO);
         _mixer->playStream(Audio::Mixer::kPlainSoundType, &_hSampleSpeech,
 stream);

         if (SwordEngine::_systemVars.useWindowsAudioMode) {
 @@ -1315,7 +1319,7 @@ void Sound::serveSample(Common::File *file, int32 i)
 {
                         if (_musicStreamFormat[i] != MusWav)
                                 len *= 2;

 -                       _musicOutputStream[i]->queueBuffer(buf, len,
 DisposeAfterUse::YES, Audio::FLAG_16BITS | Audio::FLAG_LITTLE_ENDIAN);
 +                       _musicOutputStream[i]->queueBuffer(buf, len,
 DisposeAfterUse::YES, AUDIO_FLAGS);
                 }
         }
  }
 }}}

 ...then it seems to unbreak that Vorbis release on big-endian.

 But then, it breaks some of the demos which worked before this change.

 So this could mean that the problem indeed lies somewhere around the
 different file formats/releases we support for this game.

 I'll ping Andy in our team to see if we can work from this.
-- 
Ticket URL: <https://bugs.scummvm.org/ticket/16393#comment:1>
ScummVM :: Bugs <https://bugs.scummvm.org>
ScummVM