FMOD will copy the sound buffer into its own memory when FMOD_CREATESAMPLE is used, so we don't need to keep the loaded sound files in Hunk memory anymore.

engine/Quake/snd_fmod.c

@@ -614,6 +614,7 @@ sfxcache_t *S_LoadSound(sfx_t *s)
 {
 	char namebuffer[256];
 	byte *data;
+	byte stackbuf[1 * 1024]; // avoid dirtying the cache heap
 	wavinfo_t info;
 	FMOD_CREATESOUNDEXINFO exinfo;
 	FMOD_RESULT result;
@@ -633,7 +634,7 @@ sfxcache_t *S_LoadSound(sfx_t *s)
 	q_strlcpy(namebuffer, "sound/", sizeof(namebuffer));
 	q_strlcat(namebuffer, s->name, sizeof(namebuffer));
 
-	data = COM_LoadHunkFile(namebuffer, NULL);
+	data = COM_LoadStackFile(namebuffer, stackbuf, sizeof(stackbuf), NULL);
 	if (!data)
 	{
 		Con_Printf("Couldn't load %s\n", namebuffer);
@@ -651,7 +652,8 @@ sfxcache_t *S_LoadSound(sfx_t *s)
 	exinfo.cbsize = sizeof(FMOD_CREATESOUNDEXINFO);
 	exinfo.length = com_filesize;
 
-	result = FMOD_System_CreateSound(fmod_system, (const char*)data, FMOD_3D | FMOD_OPENMEMORY, &exinfo, &s->sound);
+	// This will copy the sound data into FMOD's internal buffers, so there's no need to keep it around in hunk memory
+	result = FMOD_System_CreateSound(fmod_system, (const char*)data, FMOD_3D | FMOD_OPENMEMORY | FMOD_CREATESAMPLE, &exinfo, &s->sound);
 	if (result != FMOD_OK)
 	{
 		Con_Printf("Failed to create FMOD sound: %s\n", FMOD_ErrorString(result));