cyomtlib

Python bindings for Open Media Transport (OMT).

cyomtlib wraps libomt, the C interface to OMT, with ctypes. Encoding and decoding run in native code and NumPy arrays are passed to libomt without copying, so 1080p60 is realistic from Python. Wheels bundle libomt and the libvmx codec, so there is nothing else to install.

The package is pure Python, so one wheel per platform covers every Python 3.9+.

On Linux, discovery uses Avahi: install libavahi-client3 (Debian/Ubuntu) or avahi-libs (Fedora/RHEL) and make sure avahi-daemon is running. Without it, senders and receivers still work but sources are not discovered: connect to omt://host:port URLs or use a discovery server (omt.settings.set_discovery_server(...)). The Linux libraries carry a small patch for this (patches/); unpatched upstream libomt aborts the process when avahi-daemon is not running.

Install

pip install cyomtlib

Send

import numpy as np
import cyomtlib as omt

frame = np.zeros((1080, 1920, 4), dtype=np.uint8)  # BGRA
frame[..., 2] = 255                                  # red
frame[..., 3] = 255

with omt.Sender("Python Test Pattern") as sender:
    print("Publishing as", sender.address)
    while True:
        # timestamp=-1 (the default) lets OMT pace the frames to frame_rate.
        sender.send_video(frame, omt.Codec.BGRA, frame_rate=60)

send_video takes packed formats (BGRA, UYVY, YUY2) as arrays shaped (height, width, bytes_per_pixel) and infers the size. Planar formats (NV12, YV12, UYVA, P216, PA16) need width= and height=. Pass flags=omt.VideoFlags.ALPHA to keep an alpha channel.

Audio is planar float32, shaped (channels, samples_per_channel):

sender.send_audio(np.zeros((2, 960), np.float32), sample_rate=48000)
sender.send_metadata("<scene name='intro'/>")

Discover and receive

import cyomtlib as omt

sources = omt.discover(wait=2.0)          # ["HOSTNAME (Python Test Pattern)", ...]

with omt.Receiver(sources[0], video_format=omt.PreferredVideoFormat.BGRA) as receiver:
    while True:
        frame = receiver.receive(timeout_ms=1000)
        if isinstance(frame, omt.VideoFrame):
            pixels = frame.packed()           # (height, width, 4) uint8 view
        elif isinstance(frame, omt.AudioFrame):
            samples = frame.data              # (channels, samples) float32
        elif isinstance(frame, omt.MetadataFrame):
            print(frame.xml)

receive copies frame data into NumPy arrays you own. Pass copy=False to get views into libomt's buffers instead; they stay valid only until the next receive call for that frame type. To receive video and audio on separate threads, call receive(omt.FrameType.VIDEO) in one and receive(omt.FrameType.AUDIO) in the other.

Also available: tally (set_tally, get_tally), sender information, per-connection metadata, redirects, quality suggestions, video and audio statistics, settings, and logging:

omt.settings.log_to_python()          # OMT log lines go to logging.getLogger("cyomtlib.omt")
omt.settings.set_integer("NetworkPortStart", 7000)

Call omt.settings.shutdown() once before the process exits, after every Sender and Receiver has been closed.

Using your own libomt build

Set CYOMTLIB_LIB_DIR to a folder containing libomt and libvmx. This takes precedence over the bundled libraries. The source distribution builds a wheel without native libraries, which finds libomt this way or on the system library path.

Building wheels

python scripts/fetch_native.py          # macOS / Windows: stage the official binaries
python -m build --wheel

On Linux, build the libraries from source inside a manylinux container, then stage them:

docker run --rm -v "$PWD":/io -w /io quay.io/pypa/manylinux_2_28_x86_64 scripts/build_linux_native.sh
python scripts/fetch_native.py --from-dir build/linux-native --platform-tag manylinux_2_28_x86_64
python -m build --wheel

Upstream versions are pinned in native-sources.env. The GitHub workflow in .github/workflows/wheels.yml builds and tests all five wheels and publishes them to PyPI when a v* tag is pushed.

License

cyomtlib is MIT licensed. The bundled OMT libraries are MIT licensed by the Open Media Transport contributors (see LICENSE-OMT.txt).