monkeyplug 2.1.8

monkeyplug is a little script to censor profanity in audio files.

monkeyplug

monkeyplug is a little script to censor profanity in audio files (intended for podcasts, but YMMV) in a few simple steps:

1. The user provides a local audio file (or a URL pointing to an audio file which is downloaded)
2. Either Whisper (GitHub) or the Vosk-API is used to recognize speech in the audio file
3. Each recognized word is checked against a list of profanity or other words you'd like muted (supports text or JSON format)
4. ffmpeg is used to create a cleaned audio file, muting or "bleeping" the objectional words

You can then use your favorite media player to play the cleaned audio file.

If provided a video file for input, monkeyplug will attempt to process the audio stream from the file and remultiplex it, copying the original video stream.

monkeyplug is part of a family of projects with similar goals:

• 📼 cleanvid for video files (using SRT-formatted subtitles)
• 🎤 monkeyplug for audio and video files (using either Whisper or the Vosk-API for speech recognition)
• 📕 montag for ebooks

Installation

Using pip, to install the latest release from PyPI:

python3 -m pip install -U monkeyplug

Or to install directly from GitHub:

python3 -m pip install -U 'git+https://github.com/mmguero/monkeyplug'

Prerequisites

monkeyplug requires:

• FFmpeg
• Python 3
  • mutagen
  • a speech recognition library, either of:
    • Whisper
    • vosk-api with a VOSK compatible model

To install FFmpeg, use your operating system's package manager or install binaries from ffmpeg.org. The Python dependencies will be installed automatically if you are using pip to install monkeyplug, except for vosk or openai-whisper; as monkeyplug can work with both speech recognition engines, there is not a hard installation requirement for either until runtime.

usage

usage: monkeyplug <arguments>

options:
  -h, --help            show this help message and exit
  -v [true|false], --verbose [true|false]
                        Verbose/debug output
  -m <string>, --mode <string>
                        Speech recognition engine (whisper|vosk) (default: whisper)
  -i <string>, --input <string>
                        Input file (or URL)
  -o <string>, --output <string>
                        Output file
  --output-json <string>
                        Output file to store transcript JSON
  -w <profanity file>, --swears <profanity file>
                        text or JSON file containing profanity (default: "swears.txt")
  -a <str>, --audio-params <str>
                        Audio parameters for ffmpeg (default depends on output audio codec)
  -c <int>, --channels <int>
                        Audio output channels (default: 2)
  -s <int>, --sample-rate <int>
                        Audio output sample rate (default: 48000)
  -r <str>, --bitrate <str>
                        Audio output bitrate (default: 256K)
  -q <int>, --vorbis-qscale <int>
                        qscale for libvorbis output (default: 5)
  -f <string>, --format <string>
                        Output file format (default: inferred from extension of --output, or "MATCH")
  --pad-milliseconds <int>
                        Milliseconds to pad on either side of muted segments (default: 0)
  --pad-milliseconds-pre <int>
                        Milliseconds to pad before muted segments (default: 0)
  --pad-milliseconds-post <int>
                        Milliseconds to pad after muted segments (default: 0)
  -b [true|false], --beep [true|false]
                        Beep instead of silence
  -z <int>, --beep-hertz <int>
                        Beep frequency hertz (default: 1000)
  --beep-mix-normalize [true|false]
                        Normalize mix of audio and beeps (default: False)
  --beep-audio-weight <int>
                        Mix weight for non-beeped audio (default: 1)
  --beep-sine-weight <int>
                        Mix weight for beep (default: 1)
  --beep-dropout-transition <int>
                        Dropout transition for beep (default: 0)
  --force [true|false]  Process file despite existence of embedded tag

VOSK Options:
  --vosk-model-dir <string>
                        VOSK model directory (default: ~/.cache/vosk)
  --vosk-read-frames-chunk <int>
                        WAV frame chunk (default: 8000)

Whisper Options:
  --whisper-model-dir <string>
                        Whisper model directory (~/.cache/whisper)
  --whisper-model-name <string>
                        Whisper model name (base.en)
  --torch-threads <int>
                        Number of threads used by torch for CPU inference (0)

Docker

Alternately, a Dockerfile is provided to allow you to run monkeyplug in Docker. You can pull one of the following images:

• VOSK
  • oci.guero.org/monkeyplug:vosk-small
  • oci.guero.org/monkeyplug:vosk-large
• Whisper
  • oci.guero.org/monkeyplug:whisper-tiny.en
  • oci.guero.org/monkeyplug:whisper-tiny
  • oci.guero.org/monkeyplug:whisper-base.en
  • oci.guero.org/monkeyplug:whisper-base
  • oci.guero.org/monkeyplug:whisper-small.en
  • oci.guero.org/monkeyplug:whisper-small
  • oci.guero.org/monkeyplug:whisper-medium.en
  • oci.guero.org/monkeyplug:whisper-medium
  • oci.guero.org/monkeyplug:whisper-large-v1
  • oci.guero.org/monkeyplug:whisper-large-v2
  • oci.guero.org/monkeyplug:whisper-large-v3
  • oci.guero.org/monkeyplug:whisper-large

then run monkeyplug-docker.sh inside the directory where your audio files are located.

Contributing

If you'd like to help improve monkeyplug, pull requests will be welcomed!

Authors

• Seth Grover - Initial work - mmguero

License

This project is licensed under the BSD 3-Clause License - see the LICENSE file for details.