Python process TypeError: no default __reduce__ due to non-trivial __cinit__

Question

from aiohttp import web
from aiortc.mediastreams import MediaStreamTrack
from aiortc import RTCPeerConnection, RTCSessionDescription
from aiortc.contrib.media import MediaPlayer
from pydub import AudioSegment
import av
from aiohttp import web
from aiortc.mediastreams import MediaStreamTrack
from aiortc import RTCPeerConnection, RTCSessionDescription
from aiortc.contrib.media import MediaPlayer
from pydub import AudioSegment
import av
import pyaudio
import asyncio
import json
import os
from multiprocessing import Process, freeze_support, Queue
import sys
import threading
from time import sleep
import fractions
import time

class RadioServer(Process):
    def __init__(self,q):
        super().__init__()
        self.q = q
        self.ROOT = os.path.dirname(__file__)
        self.pcs = []
        self.channels = []
        self.stream_offers = []
    
    def run(self):
        self.app = web.Application()
        self.app.on_shutdown.append(self.on_shutdown)
        self.app.router.add_get("/", self.index)
        self.app.router.add_get("/telephone_calls.js", self.javascript)
        self.app.router.add_get("/jquery-3.5.1.min.js", self.jquery)
        self.app.router.add_post("/offer", self.offer)
        
        threading.Thread(target=self.fill_the_queues).start()
        web.run_app(self.app, access_log=None, host="192.168.1.20", port="8080", ssl_context=None)

    def fill_the_queues(self):
        while(True):
            frame = self.q.get()
            for stream_offer in self.stream_offers:
                stream_offer.q.put(frame)

    async def index(self,request):
        content = open(os.path.join(self.ROOT, "index.html"), encoding="utf8").read()
        return web.Response(content_type="text/html", text=content)


    async def javascript(self,request):
        content = open(os.path.join(self.ROOT, "telephone_calls.js"), encoding="utf8").read()
        return web.Response(content_type="application/javascript", text=content)

    async def jquery(self,request):
        content = open(os.path.join(self.ROOT, "jquery-3.5.1.min.js"), encoding="utf8").read()
        return web.Response(content_type="application/javascript", text=content)

    async def offer(self,request):  
        params = await request.json()
        offer = RTCSessionDescription(sdp=params["sdp"], type=params["type"])

        pc = RTCPeerConnection()
        self.pcs.append(pc)

        # prepare epalxeis media
        self.stream_offers.append(CustomRadioStream())
        pc.addTrack(self.stream_offers[-1])

        #player = MediaPlayer(os.path.join(self.ROOT, "ΑΓΙΑ ΚΥΡΙΑΚΗ.mp3"))
        #pc.addTrack(player.audio)

        @pc.on("datachannel")
        def on_datachannel(channel):
            self.channels.append(channel)
            self.send_channel_message(str(len(self.pcs)))


        @pc.on("iceconnectionstatechange")
        async def on_iceconnectionstatechange():
            if pc.iceConnectionState == "failed":
                self.pcs.remove(pc)
                print("Current peer connections:"+str(len(self.pcs)))
            

        # handle offer
        await pc.setRemoteDescription(offer)

        # send answer
        answer = await pc.createAnswer()
        await pc.setLocalDescription(answer)
        
            

        return web.Response(content_type="application/json",text=json.dumps({"sdp": pc.localDescription.sdp, "type": pc.localDescription.type}))

    async def on_shutdown(self,app):
        # close peer connections
        if self.pcs:
            coros = [pc.close() for pc in self.pcs]
            await asyncio.gather(*coros)
            self.pcs = []
            self.channels = []
            self.stream_offers = []
            
    def send_channel_message(self,message):
        for channel in self.channels:
            channel.send(message)
            

class CustomRadioStream(MediaStreamTrack):
    kind = "audio"
    
    def __init__(self):
        super().__init__()  # don't forget this!
        
        self.q = Queue()
        self._start = None
        
    async def recv(self):
        frame = self.q.get()
        frame_time = frame.time
        if self._start is None:
            self._start = time.time() - frame_time
        else:
            wait = self._start + frame_time - time.time()
            await asyncio.sleep(wait)
        return frame

class RadioOutputStream:
    def __init__(self,q):
        self.sample_rate = 44800
        self.AUDIO_PTIME = 0.744
        self.samples = int(self.AUDIO_PTIME * self.sample_rate)
        self.packet_time = 20

        self.FORMAT = pyaudio.paInt16
        self.CHANNELS = 2
        self.RATE = self.sample_rate
        self.CHUNK = int(44100*0.744)

        
        self.files_directory = os.path.abspath(r"C:\Users\Χρήστος\Music\Αναστάσιμα τροπάρια ή άλλα τροπάρια Δεσποτικών, Θεομητορικών ή άλλων εορτών Αγίων")
        self.files_paths = os.listdir(self.files_directory)
        self.files_info = []
        for song_file in self.files_paths:
            if ".mp3" in song_file.lower():
                file_segment = AudioSegment.from_file(os.path.join(self.files_directory, song_file)).set_frame_rate(self.sample_rate)
                duration_milliseconds = len(file_segment)
                self.files_info.append({"file_segment":file_segment,"duration_milliseconds":duration_milliseconds})
        self.total_files = len(self.files_info)
        self.current_file = 0
        self.chunk_number = 0

        self.silence = AudioSegment.silent(duration=self.packet_time)

        self.q = q

        self.codec = av.CodecContext.create('pcm_s16le', 'r')
        self.codec.sample_rate = 44800
        self.codec.channels = 2

        self.audio_samples = 0
        
    def run_stream(self):
        while(True):
            if((self.chunk_number+1)*(self.packet_time)<=self.files_info[self.current_file]["duration_milliseconds"]):
                final_slice = self.files_info[self.current_file]["file_segment"][self.chunk_number*self.packet_time:(self.chunk_number+1)*self.packet_time]
                #final_slice = final_slice + 100
                self.chunk_number += 1

                packet = av.Packet(final_slice.raw_data)
                frame = self.codec.decode(packet)[0]
                frame.pts = self.audio_samples
                frame.time_base = fractions.Fraction(1, self.codec.sample_rate)
                self.audio_samples += frame.samples
                self.q.put(frame)
            else:
                if(self.chunk_number*self.packet_time<self.files_info[self.current_file]["duration_milliseconds"]):
                    final_slice = self.files_info[self.current_file]["file_segment"][self.chunk_number*self.packet_time:]
                    final_slice = final_slice + self.silence
                    #final_slice = final_slice + 100
                    self.chunk_number += 1


                    packet = av.Packet(final_slice.raw_data)
                    frame = self.codec.decode(packet)[0]
                    frame.pts = self.audio_samples
                    frame.time_base = fractions.Fraction(1, self.codec.sample_rate)
                    self.audio_samples += frame.samples
                    self.q.put(frame)
                else:
                    #start song from begin
                    self.chunk_number=0
                    if self.current_file==self.total_files-1:
                        self.current_file = 0
                    else:
                        self.current_file += 1
                    
            sleep(0.01)

if __name__ == "__main__":
    freeze_support()

    q = Queue()
    radio_stream = RadioOutputStream(q)
    threading.Thread(target=radio_stream.run_stream).start()

    custom_server_child_process = RadioServer(q)
    custom_server_child_process.start()
    print("Thread and process started sucessfully.")

Error:

Traceback (most recent call last):
  File "C:\Python\lib\multiprocessing\queues.py", line 245, in _feed
    obj = _ForkingPickler.dumps(obj)
  File "C:\Python\lib\multiprocessing\reduction.py", line 51, in dumps
    cls(buf, protocol).dump(obj)
  File "stringsource", line 2, in av.audio.frame.AudioFrame.__reduce_cython__
TypeError: no default __reduce__ due to non-trivial __cinit__

The above error is repeated a lot of times.

What I am doing wrong?

Answer 1

I started receiving this error from OpenAI in errors returned from their whisper endpoint. Seems that I was not sending wav files formatted correctly.

I just needed to put the correct RIFF header and things started working.

Answer 2

I found it!

from aiohttp import web
from aiortc.mediastreams import MediaStreamTrack
from aiortc import RTCPeerConnection, RTCSessionDescription
from aiortc.contrib.media import MediaPlayer
from pydub import AudioSegment
import av
from aiohttp import web
from aiortc.mediastreams import MediaStreamTrack
from aiortc import RTCPeerConnection, RTCSessionDescription
from aiortc.contrib.media import MediaPlayer
from pydub import AudioSegment
import av
import pyaudio
import asyncio
import json
import os
from multiprocessing import Process, freeze_support, Queue
import sys
import threading
from time import sleep
import fractions
import time
import queue

class RadioServer(Process):
    def __init__(self,q):
        super().__init__()
        self.q = q
        self.ROOT = os.path.dirname(__file__)
        self.pcs = []
        self.channels = []
        self.stream_offers = []

    
    def run(self):
        self.codec = av.CodecContext.create('pcm_s16le', 'r')
        self.codec.sample_rate = 44800
        self.codec.channels = 2

        self.audio_samples = 0
    
        self.app = web.Application()
        self.app.on_shutdown.append(self.on_shutdown)
        self.app.router.add_get("/", self.index)
        self.app.router.add_get("/radio.js", self.javascript)
        self.app.router.add_get("/jquery-3.5.1.min.js", self.jquery)
        self.app.router.add_post("/offer", self.offer)

        threading.Thread(target=self.fill_the_queues).start()
        web.run_app(self.app, access_log=None, host="192.168.1.20", port="8080", ssl_context=None)

    def fill_the_queues(self):
        while(True):
            data = self.q.get()
            packet = av.Packet(data)
            frame = self.codec.decode(packet)[0]
            frame.pts = self.audio_samples
            frame.time_base = fractions.Fraction(1, self.codec.sample_rate)
            self.audio_samples += frame.samples
            for stream_offer in self.stream_offers:
                stream_offer.q.put(frame)

    async def index(self,request):
        content = open(os.path.join(self.ROOT, "index.html"), encoding="utf8").read()
        return web.Response(content_type="text/html", text=content)


    async def javascript(self,request):
        content = open(os.path.join(self.ROOT, "radio.js"), encoding="utf8").read()
        return web.Response(content_type="application/javascript", text=content)

    async def jquery(self,request):
        content = open(os.path.join(self.ROOT, "jquery-3.5.1.min.js"), encoding="utf8").read()
        return web.Response(content_type="application/javascript", text=content)

    async def offer(self,request):  
        params = await request.json()
        offer = RTCSessionDescription(sdp=params["sdp"], type=params["type"])

        pc = RTCPeerConnection()
        self.pcs.append(pc)

        # prepare epalxeis media
        self.stream_offers.append(CustomRadioStream())
        pc.addTrack(self.stream_offers[-1])

        #player = MediaPlayer(os.path.join(self.ROOT, "ΑΓΙΑ ΚΥΡΙΑΚΗ.mp3"))
        #pc.addTrack(player.audio)

        @pc.on("datachannel")
        def on_datachannel(channel):
            self.channels.append(channel)
            self.send_channel_message(str(len(self.pcs)))


        @pc.on("iceconnectionstatechange")
        async def on_iceconnectionstatechange():
            if pc.iceConnectionState == "failed":
                self.pcs.remove(pc)
                print("Current peer connections:"+str(len(self.pcs)))
            

        # handle offer
        await pc.setRemoteDescription(offer)

        # send answer
        answer = await pc.createAnswer()
        await pc.setLocalDescription(answer)



        return web.Response(content_type="application/json",text=json.dumps({"sdp": pc.localDescription.sdp, "type": pc.localDescription.type}))

    async def on_shutdown(self,app):
        # close peer connections
        if self.pcs:
            coros = [pc.close() for pc in self.pcs]
            await asyncio.gather(*coros)
            self.pcs = []
            self.channels = []
            self.stream_offers = []
            
    def send_channel_message(self,message):
        for channel in self.channels:
            channel.send(message)
            
class CustomRadioStream(MediaStreamTrack):
    kind = "audio"
    
    def __init__(self):
        super().__init__()  # don't forget this!
        
        self.q = queue.Queue()
        self._start = None
        
    async def recv(self):
        frame = self.q.get()
        frame_time = frame.time
        if self._start is None:
            self._start = time.time() - frame_time
        else:
            wait = self._start + frame_time - time.time()
            await asyncio.sleep(wait)
        return frame

class RadioOutputStream:
    def __init__(self,q):
        self.sample_rate = 44800
        self.AUDIO_PTIME = 0.744
        self.samples = int(self.AUDIO_PTIME * self.sample_rate)
        self.packet_time = 20

        self.FORMAT = pyaudio.paInt16
        self.CHANNELS = 2
        self.RATE = self.sample_rate
        self.CHUNK = int(44100*0.744)

        
        self.files_directory = os.path.abspath(r"C:\Users\Χρήστος\Music\Αναστάσιμα τροπάρια ή άλλα τροπάρια Δεσποτικών, Θεομητορικών ή άλλων εορτών Αγίων")
        self.files_paths = os.listdir(self.files_directory)
        self.files_info = []
        for song_file in self.files_paths:
            if ".mp3" in song_file.lower():
                file_segment = AudioSegment.from_file(os.path.join(self.files_directory, song_file)).set_frame_rate(self.sample_rate)
                duration_milliseconds = len(file_segment)
                self.files_info.append({"file_segment":file_segment,"duration_milliseconds":duration_milliseconds})
        self.total_files = len(self.files_info)
        self.current_file = 0
        self.chunk_number = 0

        self.silence = AudioSegment.silent(duration=self.packet_time)

        self.q = q
        
    def run_stream(self):
        while(True):
            if((self.chunk_number+1)*(self.packet_time)<=self.files_info[self.current_file]["duration_milliseconds"]):
                final_slice = self.files_info[self.current_file]["file_segment"][self.chunk_number*self.packet_time:(self.chunk_number+1)*self.packet_time]
                #final_slice = final_slice + 100
                self.chunk_number += 1
                self.q.put(final_slice.raw_data)
            else:
                if(self.chunk_number*self.packet_time<self.files_info[self.current_file]["duration_milliseconds"]):
                    final_slice = self.files_info[self.current_file]["file_segment"][self.chunk_number*self.packet_time:]
                    final_slice = final_slice + self.silence
                    #final_slice = final_slice + 100
                    self.chunk_number += 1


                    self.q.put(final_slice.raw_data)
                else:
                    #start song from begin
                    self.chunk_number=0
                    if self.current_file==self.total_files-1:
                        self.current_file = 0
                    else:
                        self.current_file += 1
                    
            sleep(0.01)

if __name__ == "__main__":
    freeze_support()

    q = Queue()
    radio_stream = RadioOutputStream(q)
    threading.Thread(target=radio_stream.run_stream).start()

    custom_server_child_process = RadioServer(q)
    custom_server_child_process.start()
    print("Thread and process started sucessfully.")

I "transfer" packet encoding to process, and put encoder definition inside run method.

Anyway, thanks all for your help.

Answer 3

Just a wild stab since I've never used PyAV and don't have it installed, but following from my comments, and the PyAV documentation, the AudioFrame class has from_ndarray and to_ndarray methods.

So when you could do is instead of sending the AudioFrame instances directly over the queue, you could send:

frame_data = (frame.to_ndarray(), frame.format, frame.layout)

(assuming the AudioFormat and AudioLayout classes can be pickled; if not you have to do further deconstruction).

Then send these over the queue:

self.q.put(frame_data)

When, at the other end of the queue (e.g. in CustomRadioStream.recv()) you need to reconstruct the AudioFrame object:

frame_data = self.q.get()
frame = AudioFrame.from_ndarray(frame_data[0], format=frame_data[1], layout=frame_data[2])

You could also use a namedtuple for frame_data or wrapper class to make debugging a bit easier.