183 lines
6.5 KiB
Python
183 lines
6.5 KiB
Python
from ctypes import memmove, byref, c_uint32, sizeof, cast, c_void_p, create_string_buffer, POINTER, c_char, \
|
|
c_long
|
|
|
|
from pyglet.libs.darwin import cf, CFSTR
|
|
from pyglet.libs.darwin.coreaudio import kCFURLPOSIXPathStyle, AudioStreamBasicDescription, ca, ExtAudioFileRef, \
|
|
kExtAudioFileProperty_FileDataFormat, kAudioFormatLinearPCM, kAudioFormatFlagIsSignedInteger, \
|
|
kAudioFormatFlagIsPacked, kExtAudioFileProperty_ClientDataFormat, AudioFile_ReadProc, \
|
|
AudioFile_GetSizeProc, AudioBufferList, kExtAudioFileProperty_FileLengthFrames, AudioFileID, err_check
|
|
from pyglet.media import StreamingSource, StaticSource
|
|
from pyglet.media.codecs import AudioFormat, MediaDecoder, AudioData
|
|
|
|
|
|
class MemoryFileObject:
|
|
|
|
def __init__(self, file):
|
|
self.file = file
|
|
|
|
if not getattr(self.file, 'seek', None) or not getattr(self.file, 'tell', None):
|
|
raise Exception("File object does not support seeking.")
|
|
|
|
# Seek to end of file to get the filesize.
|
|
self.file.seek(0, 2)
|
|
self.file_size = self.file.tell()
|
|
self.file.seek(0) # Put cursor back at the beginning.
|
|
self.data = []
|
|
|
|
def read_data_cb(ref, offset, requested_length, buffer, actual_count):
|
|
self.file.seek(offset)
|
|
data = self.file.read(requested_length)
|
|
data_size = len(data)
|
|
memmove(buffer, data, data_size)
|
|
actual_count.contents.value = data_size # Actual read.
|
|
return 0
|
|
|
|
def getsize_cb(ref):
|
|
return self.file_size
|
|
|
|
self.getsize_func = AudioFile_GetSizeProc(getsize_cb)
|
|
self.read_func = AudioFile_ReadProc(read_data_cb)
|
|
|
|
|
|
class CoreAudioSource(StreamingSource):
|
|
def __init__(self, filename, file=None):
|
|
self._bl = None
|
|
self._file = file
|
|
self._deleted = False
|
|
self._file_obj = None
|
|
self._audfile = None
|
|
self._audref = None
|
|
|
|
audref = ExtAudioFileRef()
|
|
if file is None:
|
|
fn_str = CFSTR(filename)
|
|
url_ref = cf.CFURLCreateWithFileSystemPath(None, fn_str, kCFURLPOSIXPathStyle, False)
|
|
|
|
err_check(ca.ExtAudioFileOpenURL(url_ref, byref(audref)))
|
|
else:
|
|
self.file_obj = MemoryFileObject(file)
|
|
|
|
self._audfile = AudioFileID()
|
|
|
|
err_check(ca.AudioFileOpenWithCallbacks(
|
|
None, self.file_obj.read_func, None, self.file_obj.getsize_func, None,
|
|
0,
|
|
byref(self._audfile))
|
|
)
|
|
|
|
err_check(ca.ExtAudioFileWrapAudioFileID(self._audfile, False, byref(audref)))
|
|
|
|
self._audref = audref
|
|
|
|
format_info = AudioStreamBasicDescription()
|
|
size = c_uint32(sizeof(format_info))
|
|
err_check(ca.ExtAudioFileGetProperty(self._audref,
|
|
kExtAudioFileProperty_FileDataFormat,
|
|
byref(size),
|
|
byref(format_info)))
|
|
|
|
self.convert_desc = self.convert_format(format_info)
|
|
|
|
err_check(ca.ExtAudioFileSetProperty(
|
|
self._audref,
|
|
kExtAudioFileProperty_ClientDataFormat,
|
|
sizeof(self.convert_desc),
|
|
byref(self.convert_desc)
|
|
))
|
|
|
|
length = c_long()
|
|
size = c_uint32(sizeof(format_info))
|
|
# File length.
|
|
err_check(ca.ExtAudioFileGetProperty(
|
|
self._audref,
|
|
kExtAudioFileProperty_FileLengthFrames,
|
|
byref(size),
|
|
byref(length)
|
|
))
|
|
|
|
self.audio_format = AudioFormat(channels=self.convert_desc.mChannelsPerFrame,
|
|
sample_size=self.convert_desc.mBitsPerChannel,
|
|
sample_rate=int(self.convert_desc.mSampleRate))
|
|
|
|
self._num_frames = length.value
|
|
self._bytes_per_frame = self.convert_desc.mBytesPerFrame
|
|
self._duration = self._num_frames / self.convert_desc.mSampleRate
|
|
self._duration_per_frame = self._duration / self._num_frames
|
|
|
|
@staticmethod
|
|
def convert_format(original_desc, bitdepth=16):
|
|
adesc = AudioStreamBasicDescription()
|
|
adesc.mSampleRate = original_desc.mSampleRate
|
|
adesc.mFormatID = kAudioFormatLinearPCM
|
|
adesc.mFormatFlags = kAudioFormatFlagIsSignedInteger | kAudioFormatFlagIsPacked
|
|
adesc.mChannelsPerFrame = original_desc.mChannelsPerFrame
|
|
adesc.mBitsPerChannel = bitdepth
|
|
adesc.mBytesPerPacket = original_desc.mChannelsPerFrame * adesc.mBitsPerChannel // 8
|
|
adesc.mFramesPerPacket = 1
|
|
adesc.mBytesPerFrame = adesc.mBytesPerPacket
|
|
return adesc
|
|
|
|
def __del__(self):
|
|
if self._file:
|
|
self._file.close()
|
|
self._file = None
|
|
|
|
if self._audfile:
|
|
err_check(ca.AudioFileClose(self._audfile))
|
|
self._audfile = None
|
|
|
|
if self._audref:
|
|
err_check(ca.ExtAudioFileDispose(self._audref))
|
|
self._audref = None
|
|
|
|
def get_audio_data(self, num_bytes, compensation_time=0.0):
|
|
num_frames = c_uint32(num_bytes // self.convert_desc.mBytesPerFrame)
|
|
|
|
if not self._bl:
|
|
buffer = create_string_buffer(num_bytes)
|
|
self._bl = AudioBufferList()
|
|
self._bl.mNumberBuffers = 1
|
|
self._bl.mBuffers[0].mNumberChannels = self.convert_desc.mChannelsPerFrame
|
|
self._bl.mBuffers[0].mDataByteSize = num_bytes
|
|
self._bl.mBuffers[0].mData = cast(buffer, c_void_p)
|
|
|
|
while True:
|
|
ca.ExtAudioFileRead(self._audref, byref(num_frames), byref(self._bl))
|
|
|
|
size = self._bl.mBuffers[0].mDataByteSize
|
|
if not size:
|
|
break
|
|
|
|
data = cast(self._bl.mBuffers[0].mData, POINTER(c_char))
|
|
slice = data[:size]
|
|
return AudioData(slice, size, 0.0, 0.0, [])
|
|
|
|
def seek(self, timestamp):
|
|
timestamp = max(0.0, min(timestamp, self._duration))
|
|
position = int(timestamp / self._duration_per_frame)
|
|
ca.ExtAudioFileSeek(self._audref, position)
|
|
|
|
|
|
#########################################
|
|
# Decoder class:
|
|
#########################################
|
|
|
|
class CoreAudioDecoder(MediaDecoder):
|
|
|
|
def get_file_extensions(self):
|
|
return '.aac', '.ac3', '.aif', '.aiff', '.aifc', '.caf', '.mp3', '.mp4', '.m4a', '.snd', '.au', '.sd2', '.wav'
|
|
|
|
def decode(self, filename, file, streaming=True):
|
|
if streaming:
|
|
return CoreAudioSource(filename, file)
|
|
else:
|
|
return StaticSource(CoreAudioSource(filename, file))
|
|
|
|
|
|
def get_decoders():
|
|
return [CoreAudioDecoder()]
|
|
|
|
|
|
def get_encoders():
|
|
return []
|