Clairvoyance: Concurrent, CLI Lip-reader

Copyright (C) 2019 Ken-ya and Takahiro Yoshimura. All rights reserved. Licensed under the terms of GNU General Public License Version 3 or later. See COPYING.txt for details.

Clairvoyance is a concurrent lip reader. It is capable to read speakers are speaking in the given video stream, showing the threads of conversations.

Quickstart

You need: Python 3.6, ffmpeg (with ffprobe)

$ python3.6 -m venv ~/ve/cv
$ ~/ve/cv/bin/pip git+https://github.com/monolithworks/clairvoyance
$ ~/ve/cv/bin/pip git+https://github.com/monolithworks/clairvoyance-lipnet
$ ~/ve/cv/bin/clairvoyance --debug --show-frame --face-detect-subsample=1 /path/to/some/movie.mp4
...
DEBUG:FaceDetector:known_0: 70: recognized
DEBUG:FaceDetector:known_1: 70: recognized
DEBUG:FaceDetector:known_2: 70: recognized
DEBUG:FaceDetector:known_3: 70: recognized
...
DEBUG:FaceRecognitionTask:Faces detected: 4 (18.67 sec.).
known_0: bin re at z four now (6.60 sec)
known_1: bin bred ich c five son (3.63 sec)
known_2: ben are in f see son (3.72 sec)
known_3: h b o n o one son (6.10 sec)

Bugs

Lots; slow, arcane, have to be stopped with C-c after detection, and so on. Sorry for that.