Module livekit.agents.utils.codecs

Sub-modules

livekit.agents.utils.codecs.mp3

Classes

class Mp3StreamDecoder

A class that can be used to stream arbitrary MP3 data (i.e. from an HTTP chunk) and decode it into PCM audio. This class is meant to be ephemeral. When you're done sending data, call close() to flush the decoder and create a new instance of this class if you need to decode more data.

Expand source code
class Mp3StreamDecoder:
    """A class that can be used to stream arbitrary MP3 data (i.e. from an HTTP chunk) and decode it into PCM audio.
    This class is meant to be ephemeral. When you're done sending data, call close() to flush
    the decoder and create a new instance of this class if you need to decode more data.
    """

    def __init__(self):
        try:
            globals()["av"] = import_module("av")
        except ImportError:
            raise ImportError(
                "You haven't included the 'codecs' optional dependencies. Please install the 'codecs' extra by running `pip install livekit-agents[codecs]`"
            )

        self._codec = av.CodecContext.create("mp3", "r")  # noqa

    def decode_chunk(self, chunk: bytes) -> List[rtc.AudioFrame]:
        packets = self._codec.parse(chunk)
        result: List[rtc.AudioFrame] = []
        for packet in packets:
            try:
                decoded = self._codec.decode(packet)
            except Exception as e:
                logging.warning(f"Error decoding packet, skipping: {e}")
                continue
            for frame in decoded:
                nchannels = len(frame.layout.channels)
                if frame.format.is_planar and nchannels > 1:
                    logging.warning(
                        "TODO: planar audio has not yet been considered, skipping frame"
                    )
                    continue
                plane = frame.planes[0]
                ptr = plane.buffer_ptr
                size = plane.buffer_size
                byte_array_pointer = ctypes.cast(
                    ptr, ctypes.POINTER(ctypes.c_char * size)
                )
                result.append(
                    rtc.AudioFrame(
                        data=bytes(byte_array_pointer.contents),
                        num_channels=nchannels,
                        sample_rate=frame.sample_rate,
                        samples_per_channel=frame.samples,
                    )
                )
        return result

Methods

def decode_chunk(self, chunk: bytes) ‑> List[AudioFrame]