aio-rtsp-toolkit

aio-rtsp-toolkit is an asyncio-based RTSP toolkit for Python. It provides:

• An RTSP client that exposes connection events, RTSP method timing, RTP packets, and assembled audio/video frames
• A lightweight RTSP/TCP server that publishes local media files from a directory tree

It is aimed at cases where "can play" is not enough and you also want visibility into protocol timing, packet flow, frame boundaries.

Highlights

• One async event stream for connect, RTSP, RTP, video, audio, and close events
• Per-method timing for OPTIONS, DESCRIBE, SETUP, PLAY, and TEARDOWN
• H.264 / H.265 video frame splicing to raw Annex B NAL units
• Audio frame extraction for AAC, G.711 A-law, and G.711 mu-law style RTSP streams
• Simple embedded RTSP server for local files
• Session API that works cleanly with async with

Timing values such as session_elapsed, RTSP elapsed, and RTP recv_tick are process-local monotonic durations. They are not wall-clock timestamps.

Installation

Base install:

pip install aio-rtsp-toolkit

This installs three command-line entry points:

• rtsp-client-cli
• rtsp-client-pyqt5
• rtsp-server

On Windows they are installed as .exe launchers in the active Python Scripts directory. On Linux and macOS they are installed into the active Python bin directory.

Optional extras:

pip install aio-rtsp-toolkit[server]
pip install aio-rtsp-toolkit[decode]
pip install aio-rtsp-toolkit[audio]
pip install aio-rtsp-toolkit[all]

• [server]: installs av for RTSP file serving
• [decode]: installs numpy and av for audio PCM decode helpers
• [audio]: installs numpy, sounddevice, and av for audio playback helpers
• [all]: installs all optional dependencies

On Linux, audio playback also requires the system PortAudio runtime in addition to the Python sounddevice package. Without PortAudio, rtsp-client-pyqt5 and rtsp-client-cli --audio-mode play will fail at runtime.

On Ubuntu, install it with:

sudo apt update
sudo apt install -y libportaudio2

Quick Start

Client

If the RTSP stream requires authentication, include credentials in the URL, for example rtsp://user:password@192.168.1.122:554/stream. If the username or password contains reserved characters such as @, :, or /, URL-encode them first.

import asyncio
import aio_rtsp_toolkit as aiortsp


async def main():
    async with aiortsp.RtspSession("rtsp://127.0.0.1:8554/zhongli.wav", timeout=5) as session:
        async for event in session.iter_events():
            if isinstance(event, aiortsp.ConnectResultEvent):
                print("connected:", event.local_addr, "elapsed:", event.elapsed)
            elif isinstance(event, aiortsp.RtspMethodEvent):
                print(event.method, event.status_code, event.elapsed, event.session_elapsed)
            elif isinstance(event, aiortsp.RtpPacketEvent):
                pass
            elif isinstance(event, aiortsp.VideoFrameEvent):
                print("video ts=", event.frame.timestamp, "size=", len(event.frame.data))
            elif isinstance(event, aiortsp.AudioFrameEvent):
                print("audio ts=", event.frame.timestamp, "samples=", event.frame.sample_count)
            elif isinstance(event, aiortsp.ClosedEvent):
                print("closed after", event.session_elapsed, "seconds")


asyncio.run(main())

Server

Publish a directory recursively:

rtsp-server --dir ./media --host 0.0.0.0 --port 8554

Each file becomes an RTSP resource under the same relative path:

rtsp://127.0.0.1:8554/morning_h264.mp4
rtsp://127.0.0.1:8554/subdir/example.wav

Run the server from Python:

import asyncio
import aio_rtsp_toolkit.server as server


async def main():
    await server.serve("./media", host="0.0.0.0", port=8554)


asyncio.run(main())

RTSP Server

Current server behavior:

• Transport is RTSP over TCP with interleaved RTP/RTCP only
• Files are resolved relative to the configured root directory
• Files outside the root are rejected
• Unsupported codecs are skipped rather than transcoded
• Session timeout negotiation is not implemented
• The server sends minimal RTCP Sender Report and BYE packets for configured tracks

Supported Inputs

Accepted file extensions: .mp4, .mkv, .wav and .aac.

Supported media handling:

• Video: H.264 and H.265
• AAC audio: served as mpeg4-generic
• WAV audio: PCM WAV input is resampled with PyAV and served as PCMA/8000/1
• PCMU is not advertised by the current server implementation

Loop Control

Use the play_count query parameter:

rtsp://127.0.0.1:8554/zhongli.wav?play_count=1
rtsp://127.0.0.1:8554/zhongli.wav?play_count=2
rtsp://127.0.0.1:8554/zhongli.wav?play_count=0

• play_count=1: play once, then close
• play_count=2: play twice, then close
• play_count=0: loop forever
• play_count<0 or omitted play_count: play once, then close

RTSP Client

Session Notes

• Use enable_video=False or enable_audio=False for single-media sessions
• At least one of enable_video or enable_audio must be True
• iter_events(stop_event) accepts threading.Event or asyncio.Event
• VideoFrameEvent.frame.data contains raw Annex B video data

PyAV Decode Example

Use this pattern if you want to decode video frames after DESCRIBE:

import asyncio
import fractions

import av
import aio_rtsp_toolkit as aiortsp


async def main():
    codec = None
    time_base = fractions.Fraction(1, 90000)

    async with aiortsp.RtspSession("rtsp://127.0.0.1:8554/morning_h264.mp4", timeout=5) as session:
        async for event in session.iter_events():
            if isinstance(event, aiortsp.RtspMethodEvent) and event.method == "DESCRIBE":
                video_sdp = event.response.sdp.get("video", {})
                codec_name = video_sdp.get("codec_name", "").lower()
                if codec_name:
                    av_codec_name = aiortsp.HEVCCodecName if codec_name == aiortsp.H265CodecName else codec_name
                    codec = av.CodecContext.create(av_codec_name, "r")
                    time_base = fractions.Fraction(1, video_sdp.get("clock_rate", 90000))
                    for key in ("sps", "pps"):
                        extra = video_sdp.get(key)
                        if extra:
                            codec.parse(extra)

            elif isinstance(event, aiortsp.VideoFrameEvent) and codec is not None:
                for packet in codec.parse(event.frame.data):
                    packet.pts = packet.dts = event.frame.timestamp
                    packet.time_base = time_base
                    for decoded in codec.decode(packet):
                        print("decoded video frame pts=", decoded.pts)


asyncio.run(main())

Optional Audio Decode / Playback Helpers

aio_rtsp_toolkit.audio_decode.AudioPCMDecoder can decode audio frames to PCM s16le.

• feed() returns a list of PCMFrame objects
• PCMFrame.pcm_bytes contains interleaved PCM bytes
• Set output_sample_rate to resample to a target rate
• Leave output_sample_rate=None to keep the source sample rate

aio_rtsp_toolkit.audio_playback.SoundDeviceAudioPlayer builds on the decoder in audio_decode and plays the decoded PCM immediately.

• Install the [decode] extra for PCM decode only, or [audio] for decode + playback
• G.711 A-law and mu-law playback do not require PyAV
• AAC and AAC-LATM playback do require PyAV
• On Linux, playback also requires the system PortAudio runtime

Demos

PyQt Demo

pyqt_demo.py shows how to consume RtspSession events, decode video with PyAV, and play audio with sounddevice.

It also requires PyQt5 in addition to the [audio] extra. On Linux it also requires the system PortAudio runtime.

rtsp-client-pyqt5

CLI Demo

cli_demo.py logs RTSP timing, writes raw video to disk, and decodes frames with PyAV.

You can still change the source defaults in the script, but it now also supports runtime control:

rtsp-client-cli -u rtsp://127.0.0.1:8554/morning_h264.mp4
rtsp-client-cli -u rtsp://127.0.0.1:8554/morning_h264.mp4 --audio-mode decode --audio-rate 16000
rtsp-client-cli -u rtsp://127.0.0.1:8554/morning_h264.mp4 --save-audio ""
rtsp-client-cli -u rtsp://127.0.0.1:8554/morning_h264.mp4 --audio-mode play --audio-rate 16000 --session-id s01 --log-prefix cam01

• --audio-mode decode: decode audio to PCM
• --audio-mode play: decode audio and play it
• --audio-rate 16000: resample output PCM to 16000
• omit --audio-rate: keep the source sample rate without resampling
• --save-audio audio.wav: save decoded audio to WAV, default is audio.wav
• --save-audio "": disable audio saving
• --session-id s01: set the session id used by RTSP/audio logs
• --log-prefix cam01: set the log prefix used by RTSP/audio logs

Install the [all] extra for PyAV, then install Pillow if you want to save the first decoded frame as an image. On Linux, --audio-mode play also requires the system PortAudio runtime.

rtsp-client-cli -u rtsp://127.0.0.1:8554/morning_h264.mp4

API Summary

RtspSession(rtsp_url, forward_address=None, timeout=4, session_id=None, enable_video=True, enable_audio=True, log_type=RtspClientMsgType.RTSP, log_prefix='')

High-level reusable RTSP session object. Use it with async with and consume events through iter_events().

forward_address lets you keep the original RTSP URL while connecting through another TCP endpoint, such as a relay, tunnel, or forwarded host.

RtspSession.iter_events(stop_event=None) -> AsyncGenerator[RtspEvent, None]

Starts the session, yields typed events, and closes the socket when iteration ends or fails.

open_session(rtsp_url, forward_address=None, timeout=4, session_id=None, enable_video=True, enable_audio=True, log_type=RtspClientMsgType.RTSP, log_prefix='') -> RtspSession

Convenience helper that returns a RtspSession.

Event Types

• ConnectResultEvent
• RtspMethodEvent
• RtpPacketEvent
• VideoFrameEvent
• AudioFrameEvent
• ClosedEvent

Exceptions

• RtspError
• RtspConnectionError
• RtspTimeoutError
• RtspProtocolError
• RtspResponseError

Notes

If you only need playback, higher-level PyAV or ffmpeg-based wrappers may be simpler. This project is most useful when you need direct access to RTSP/RTP behavior, timing, and frame-level diagnostics.