RTP Audio Separation

Task Description

Implement a program that reads raw RTP packet data from a file or network interface and produces audio files.

You are provided with a PCAP file containing packets captured from an Avaya telephony system. It contains thousands of audio streams captured within a 180 second duration. Your task is to identify and write each audio stream to a separate audio file.

Although we do not require that the program run in real time, we expect you to be aware of performance considerations and attempt to optimize your program to run close to it. You are allowed to use Python, Go, C, or C++ to implement your program. You are expected to utilize concurrency to complete this task.

Your program should accept either a network interface or capture file, and output directory as input parameters. Additional configuration should be handled using optional parameters, environment variables, and/or a configuration file. For example: ./audio_rtp_parser -i eno3 -o /tmp/audio -c config.yml

You must verify that the output of your program is correct, and are expected to submit your source and testing code via a git repository. Please commit often and with adequate commit messages; how you approach and think through the problem is as important as the final outcome.

Approach

This repo contain 2 main python script, main.py and main_parallel.py. Both script using PyShark module that wrap Wireshark / tshark.

PyShark module has limitation to read the PCAP file (doing fileCapture). In my experiment, PyShark FileCapture has limitation to read packet persecond (pps) with following formula pps = 0.0021(n_packets) + 2.6299. In other words, with increasing number of packets, more time needed to read, despite stream only 180s it finish read more than 180s.

packet/time (pps)	n_packet
21.04372093	8884
41.71766881	17779
91.09865561	41318

ReadStream process is colleting payload each session, detecting if there is any pair (forward-reverse stream) audio stream, detecting codec used on each session.

So, the difference between main.py and main_parallel.py is on how to process payload extract the audio file, and if there is reverse packet (pair), it will be merge using ffmpeg library. In the main.py process extraction and merging in the same thread (main thread), in the other hand, main_parallel.py process extraction and merging in the difference thread (extraction thread and merging thread), proceed by queue of the RTP session (rtp.ssrc).

Requirements

• Wireshark / tshark

  choco install wireshark

• ffmpeg

  choco install ffmpeg

• Python 3.6+ with following modules : Cython, pyshark, pywav, pydub, and pyyaml

  pip install -r requirements.txt

Testing Enviroment

Label	Value
OS	Windows 10 Version 10.0.18363 Build 18363
Memory	16GB DDR4
Processor	4 cores Intel Core i7-8550U @1.8GHz

Configuration Sample

# config.yml
filter: rtp.ssrc==0x46873251||rtp.ssrc==0x0d730ec4

Usage

• Single Thread

  python main.py -i file.pcap -o outdir -c config.yml

• Multi Thread

  python main_parallel.py -i file.pcap -o outdir -c config.yml

Logs

If you want to see logs, check logs dir