mirror of
https://github.com/markqvist/Sideband.git
synced 2024-12-27 00:19:25 -05:00
359 lines
13 KiB
Python
359 lines
13 KiB
Python
|
import ctypes
|
||
|
from typing import Optional, Union, ByteString
|
||
|
|
||
|
from . import opus
|
||
|
from .pyogg_error import PyOggError
|
||
|
|
||
|
class OpusEncoder:
|
||
|
"""Encodes PCM data into Opus frames."""
|
||
|
def __init__(self) -> None:
|
||
|
self._encoder: Optional[ctypes.pointer] = None
|
||
|
self._channels: Optional[int] = None
|
||
|
self._samples_per_second: Optional[int] = None
|
||
|
self._application: Optional[int] = None
|
||
|
self._max_bytes_per_frame: Optional[opus.opus_int32] = None
|
||
|
self._output_buffer: Optional[ctypes.Array] = None
|
||
|
self._output_buffer_ptr: Optional[ctypes.pointer] = None
|
||
|
|
||
|
# An output buffer of 4,000 bytes is recommended in
|
||
|
# https://opus-codec.org/docs/opus_api-1.3.1/group__opus__encoder.html
|
||
|
self.set_max_bytes_per_frame(4000)
|
||
|
|
||
|
#
|
||
|
# User visible methods
|
||
|
#
|
||
|
|
||
|
def set_channels(self, n: int) -> None:
|
||
|
"""Set the number of channels.
|
||
|
|
||
|
n must be either 1 or 2.
|
||
|
|
||
|
"""
|
||
|
if self._encoder is None:
|
||
|
if n < 0 or n > 2:
|
||
|
raise PyOggError(
|
||
|
"Invalid number of channels in call to "+
|
||
|
"set_channels()"
|
||
|
)
|
||
|
self._channels = n
|
||
|
else:
|
||
|
raise PyOggError(
|
||
|
"Cannot change the number of channels after "+
|
||
|
"the encoder was created. Perhaps "+
|
||
|
"set_channels() was called after encode()?"
|
||
|
)
|
||
|
|
||
|
def set_sampling_frequency(self, samples_per_second: int) -> None:
|
||
|
"""Set the number of samples (per channel) per second.
|
||
|
|
||
|
This must be one of 8000, 12000, 16000, 24000, or 48000.
|
||
|
|
||
|
Regardless of the sampling rate and number of channels
|
||
|
selected, the Opus encoder can switch to a lower audio
|
||
|
bandwidth or number of channels if the bitrate selected is
|
||
|
too low. This also means that it is safe to always use 48
|
||
|
kHz stereo input and let the encoder optimize the
|
||
|
encoding.
|
||
|
|
||
|
"""
|
||
|
if self._encoder is None:
|
||
|
if samples_per_second in [8000, 12000, 16000, 24000, 48000]:
|
||
|
self._samples_per_second = samples_per_second
|
||
|
else:
|
||
|
raise PyOggError(
|
||
|
"Specified sampling frequency "+
|
||
|
"({:d}) ".format(samples_per_second)+
|
||
|
"was not one of the accepted values"
|
||
|
)
|
||
|
else:
|
||
|
raise PyOggError(
|
||
|
"Cannot change the sampling frequency after "+
|
||
|
"the encoder was created. Perhaps "+
|
||
|
"set_sampling_frequency() was called after encode()?"
|
||
|
)
|
||
|
|
||
|
def set_application(self, application: str) -> None:
|
||
|
"""Set the encoding mode.
|
||
|
|
||
|
This must be one of 'voip', 'audio', or 'restricted_lowdelay'.
|
||
|
|
||
|
'voip': Gives best quality at a given bitrate for voice
|
||
|
signals. It enhances the input signal by high-pass
|
||
|
filtering and emphasizing formants and
|
||
|
harmonics. Optionally it includes in-band forward error
|
||
|
correction to protect against packet loss. Use this mode
|
||
|
for typical VoIP applications. Because of the enhancement,
|
||
|
even at high bitrates the output may sound different from
|
||
|
the input.
|
||
|
|
||
|
'audio': Gives best quality at a given bitrate for most
|
||
|
non-voice signals like music. Use this mode for music and
|
||
|
mixed (music/voice) content, broadcast, and applications
|
||
|
requiring less than 15 ms of coding delay.
|
||
|
|
||
|
'restricted_lowdelay': configures low-delay mode that
|
||
|
disables the speech-optimized mode in exchange for
|
||
|
slightly reduced delay. This mode can only be set on an
|
||
|
newly initialized encoder because it changes the codec
|
||
|
delay.
|
||
|
"""
|
||
|
if self._encoder is not None:
|
||
|
raise PyOggError(
|
||
|
"Cannot change the application after "+
|
||
|
"the encoder was created. Perhaps "+
|
||
|
"set_application() was called after encode()?"
|
||
|
)
|
||
|
if application == "voip":
|
||
|
self._application = opus.OPUS_APPLICATION_VOIP
|
||
|
elif application == "audio":
|
||
|
self._application = opus.OPUS_APPLICATION_AUDIO
|
||
|
elif application == "restricted_lowdelay":
|
||
|
self._application = opus.OPUS_APPLICATION_RESTRICTED_LOWDELAY
|
||
|
else:
|
||
|
raise PyOggError(
|
||
|
"The application specification '{:s}' ".format(application)+
|
||
|
"wasn't one of the accepted values."
|
||
|
)
|
||
|
|
||
|
def set_max_bytes_per_frame(self, max_bytes: int) -> None:
|
||
|
"""Set the maximum number of bytes in an encoded frame.
|
||
|
|
||
|
Size of the output payload. This may be used to impose an
|
||
|
upper limit on the instant bitrate, but should not be used
|
||
|
as the only bitrate control.
|
||
|
|
||
|
TODO: Use OPUS_SET_BITRATE to control the bitrate.
|
||
|
|
||
|
"""
|
||
|
self._max_bytes_per_frame = opus.opus_int32(max_bytes)
|
||
|
OutputBuffer = ctypes.c_ubyte * max_bytes
|
||
|
self._output_buffer = OutputBuffer()
|
||
|
self._output_buffer_ptr = (
|
||
|
ctypes.cast(ctypes.pointer(self._output_buffer),
|
||
|
ctypes.POINTER(ctypes.c_ubyte))
|
||
|
)
|
||
|
|
||
|
|
||
|
def encode(self, pcm: Union[bytes, bytearray, memoryview]) -> memoryview:
|
||
|
"""Encodes PCM data into an Opus frame.
|
||
|
|
||
|
`pcm` must be formatted as bytes-like, with each sample taking
|
||
|
two bytes (signed 16-bit integers; interleaved left, then
|
||
|
right channels if in stereo).
|
||
|
|
||
|
If `pcm` is not writeable, a copy of the array will be made.
|
||
|
|
||
|
"""
|
||
|
# If we haven't already created an encoder, do so now
|
||
|
if self._encoder is None:
|
||
|
self._encoder = self._create_encoder()
|
||
|
|
||
|
# Sanity checks also satisfy mypy type checking
|
||
|
assert self._channels is not None
|
||
|
assert self._samples_per_second is not None
|
||
|
assert self._output_buffer is not None
|
||
|
|
||
|
# Calculate the effective frame duration of the given PCM
|
||
|
# data. Calculate it in units of 0.1ms in order to avoid
|
||
|
# floating point comparisons.
|
||
|
bytes_per_sample = 2
|
||
|
frame_size = (
|
||
|
len(pcm) # bytes
|
||
|
// bytes_per_sample
|
||
|
// self._channels
|
||
|
)
|
||
|
frame_duration = (
|
||
|
(10*frame_size)
|
||
|
// (self._samples_per_second//1000)
|
||
|
)
|
||
|
|
||
|
# Check that we have a valid frame size
|
||
|
if int(frame_duration) not in [25, 50, 100, 200, 400, 600]:
|
||
|
raise PyOggError(
|
||
|
"The effective frame duration ({:.1f} ms) "
|
||
|
.format(frame_duration/10)+
|
||
|
"was not one of the acceptable values."
|
||
|
)
|
||
|
|
||
|
# Create a ctypes object sharing the memory of the PCM data
|
||
|
PcmCtypes = ctypes.c_ubyte * len(pcm)
|
||
|
try:
|
||
|
# Attempt to share the PCM memory
|
||
|
|
||
|
# Unfortunately, as at 2020-09-27, the type hinting for
|
||
|
# read-only and writeable buffer protocols was a
|
||
|
# work-in-progress. The following only works for writable
|
||
|
# cases, but the method's parameters include a read-only
|
||
|
# possibility (bytes), thus we ignore mypy's error.
|
||
|
pcm_ctypes = PcmCtypes.from_buffer(pcm) # type: ignore[arg-type]
|
||
|
except TypeError:
|
||
|
# The data must be copied if it's not writeable
|
||
|
pcm_ctypes = PcmCtypes.from_buffer_copy(pcm)
|
||
|
|
||
|
# Create a pointer to the PCM data
|
||
|
pcm_ptr = ctypes.cast(
|
||
|
pcm_ctypes,
|
||
|
ctypes.POINTER(opus.opus_int16)
|
||
|
)
|
||
|
|
||
|
# Create an int giving the frame size per channel
|
||
|
frame_size_int = ctypes.c_int(frame_size)
|
||
|
|
||
|
# Encode PCM
|
||
|
result = opus.opus_encode(
|
||
|
self._encoder,
|
||
|
pcm_ptr,
|
||
|
frame_size_int,
|
||
|
self._output_buffer_ptr,
|
||
|
self._max_bytes_per_frame
|
||
|
)
|
||
|
|
||
|
# Check for any errors
|
||
|
if result < 0:
|
||
|
raise PyOggError(
|
||
|
"An error occurred while encoding to Opus format: "+
|
||
|
opus.opus_strerror(result).decode("utf")
|
||
|
)
|
||
|
|
||
|
# Get memoryview of buffer so that the slice operation doesn't
|
||
|
# copy the data.
|
||
|
#
|
||
|
# Unfortunately, as at 2020-09-27, the type hints for
|
||
|
# memoryview do not include ctype arrays. This is because
|
||
|
# there is no currently accepted manner to label a class as
|
||
|
# supporting the buffer protocol. However, it's clearly a
|
||
|
# work in progress. For more information, see:
|
||
|
# * https://bugs.python.org/issue27501
|
||
|
# * https://github.com/python/typing/issues/593
|
||
|
# * https://github.com/python/typeshed/pull/4232
|
||
|
mv = memoryview(self._output_buffer) # type: ignore
|
||
|
|
||
|
# Cast the memoryview to char
|
||
|
mv = mv.cast('c')
|
||
|
|
||
|
# Slice just the valid data from the memoryview
|
||
|
valid_data_as_bytes = mv[:result]
|
||
|
|
||
|
# DEBUG
|
||
|
# Convert memoryview back to ctypes instance
|
||
|
Buffer = ctypes.c_ubyte * len(valid_data_as_bytes)
|
||
|
buf = Buffer.from_buffer( valid_data_as_bytes )
|
||
|
|
||
|
# Convert PCM back to pointer and dump 4,000-byte buffer
|
||
|
ptr = ctypes.cast(
|
||
|
buf,
|
||
|
ctypes.POINTER(ctypes.c_ubyte)
|
||
|
)
|
||
|
|
||
|
return valid_data_as_bytes
|
||
|
|
||
|
|
||
|
def get_algorithmic_delay(self):
|
||
|
"""Gets the total samples of delay added by the entire codec.
|
||
|
|
||
|
This can be queried by the encoder and then the provided
|
||
|
number of samples can be skipped on from the start of the
|
||
|
decoder's output to provide time aligned input and
|
||
|
output. From the perspective of a decoding application the
|
||
|
real data begins this many samples late.
|
||
|
|
||
|
The decoder contribution to this delay is identical for all
|
||
|
decoders, but the encoder portion of the delay may vary from
|
||
|
implementation to implementation, version to version, or even
|
||
|
depend on the encoder's initial configuration. Applications
|
||
|
needing delay compensation should call this method rather than
|
||
|
hard-coding a value.
|
||
|
|
||
|
"""
|
||
|
# If we haven't already created an encoder, do so now
|
||
|
if self._encoder is None:
|
||
|
self._encoder = self._create_encoder()
|
||
|
|
||
|
# Obtain the algorithmic delay of the Opus encoder. See
|
||
|
# https://tools.ietf.org/html/rfc7845#page-27
|
||
|
delay = opus.opus_int32()
|
||
|
|
||
|
result = opus.opus_encoder_ctl(
|
||
|
self._encoder,
|
||
|
opus.OPUS_GET_LOOKAHEAD_REQUEST,
|
||
|
ctypes.pointer(delay)
|
||
|
)
|
||
|
if result != opus.OPUS_OK:
|
||
|
raise PyOggError(
|
||
|
"Failed to obtain the algorithmic delay of "+
|
||
|
"the Opus encoder: "+
|
||
|
opus.opus_strerror(result).decode("utf")
|
||
|
)
|
||
|
delay_samples = delay.value
|
||
|
return delay_samples
|
||
|
|
||
|
|
||
|
#
|
||
|
# Internal methods
|
||
|
#
|
||
|
|
||
|
def _create_encoder(self) -> ctypes.pointer:
|
||
|
# To create an encoder, we must first allocate resources for it.
|
||
|
# We want Python to be responsible for the memory deallocation,
|
||
|
# and thus Python must be responsible for the initial memory
|
||
|
# allocation.
|
||
|
|
||
|
# Check that the application has been defined
|
||
|
if self._application is None:
|
||
|
raise PyOggError(
|
||
|
"The application was not specified before "+
|
||
|
"attempting to create an Opus encoder. Perhaps "+
|
||
|
"encode() was called before set_application()?"
|
||
|
)
|
||
|
application = self._application
|
||
|
|
||
|
# Check that the sampling frequency has been defined
|
||
|
if self._samples_per_second is None:
|
||
|
raise PyOggError(
|
||
|
"The sampling frequency was not specified before "+
|
||
|
"attempting to create an Opus encoder. Perhaps "+
|
||
|
"encode() was called before set_sampling_frequency()?"
|
||
|
)
|
||
|
|
||
|
# The frequency must be passed in as a 32-bit int
|
||
|
samples_per_second = opus.opus_int32(self._samples_per_second)
|
||
|
|
||
|
# Check that the number of channels has been defined
|
||
|
if self._channels is None:
|
||
|
raise PyOggError(
|
||
|
"The number of channels were not specified before "+
|
||
|
"attempting to create an Opus encoder. Perhaps "+
|
||
|
"encode() was called before set_channels()?"
|
||
|
)
|
||
|
channels = self._channels
|
||
|
|
||
|
# Obtain the number of bytes of memory required for the encoder
|
||
|
size = opus.opus_encoder_get_size(channels);
|
||
|
|
||
|
# Allocate the required memory for the encoder
|
||
|
memory = ctypes.create_string_buffer(size)
|
||
|
|
||
|
# Cast the newly-allocated memory as a pointer to an encoder. We
|
||
|
# could also have used opus.oe_p as the pointer type, but writing
|
||
|
# it out in full may be clearer.
|
||
|
encoder = ctypes.cast(memory, ctypes.POINTER(opus.OpusEncoder))
|
||
|
|
||
|
# Initialise the encoder
|
||
|
error = opus.opus_encoder_init(
|
||
|
encoder,
|
||
|
samples_per_second,
|
||
|
channels,
|
||
|
application
|
||
|
)
|
||
|
|
||
|
# Check that there hasn't been an error when initialising the
|
||
|
# encoder
|
||
|
if error != opus.OPUS_OK:
|
||
|
raise PyOggError(
|
||
|
"An error occurred while creating the encoder: "+
|
||
|
opus.opus_strerror(error).decode("utf")
|
||
|
)
|
||
|
|
||
|
# Return our newly-created encoder
|
||
|
return encoder
|