A command-line tool that normalizes audio and reduces noise. This tool aims to effortless create an audio file that is high in quality. It targets an integrated loudness (LUFS, RMS) close to 20, by default. It uses ffmpeg-normalize and additional ffmpeg filters to achieve the desired result, as well as SoX and mp3gain.
Audiotame has an ACX Check. Notice, however, that while meeting these requirements may be desirable for some users, it not guaranteed to be achieved on the first run without some tweaks.
Also, it uses Bash, so it is not expected to work on Windows systems, unless used under WSL or containers.
A live demo is running on HuggingFace
Contents
ffmpegsoxmp3gainpython3python3-pip
python3 -m pip install audiotame
If you wish to use this program with Gradio's Web UI:
python3 -m pip install audiotame[gui]
Notice: even installing the pip package, you still need to install system packages.
git clone --recurse-submodules https://github.com/veralvx/audiotame.git audiotame
cd audiotame
podman build . -t audiotame
podman pull ghcr.io/veralvx/audiotame:cli
podman pull ghcr.io/veralvx/audiotame:gradio
Then, you can run:
podman run --rm -it --volume "$(pwd)":/workspace audiotame:cli
podman run -it -p 7860:7860 audiotame:gradio
Use an alias in your ~/.bashrc:
alias audiotame='podman run --rm --volume $(pwd):/workspace audiotame:cli'
alias audiotame:gradio='podman run -it --rm -p 7860:7860 audiotame:gradio'
echo "alias audiotame='podman run --rm --volume \$(pwd):/workspace audiotame:cli'" >> ~/.bashrc
echo "alias audiotame:gradio='podman run -it --rm -p 7860:7860 audiotame:gradio'" >> ~/.bashrc
Then,
source ~/.bashrc
The directory from wich you run audiotame command will be mounted in /workspace, which is also the working directory. So, a file named audio.wav in your current directory can be passed to the program with:
audiotame audio.wav
Usage: audiotame {path_to_file | --gradio} [operation] [operation_arg]
Flag:
--gradio Start Gradio server
Operations:
pass Do not alter peak level db
stats Display audio file statistics
acx Check for ACX compatibility
sr <sample_rate> Change sample rate (e.g., 44100)
br <bitrate> Change bitrate (e.g., 128k, 320k)
convert <format> Convert to specified format (e.g., mp3, wav)
extract Extract audio from video
Example:
audiotame path_to_file
This outputs file-tamed.wav in the same directory as the file is located.
Notice: The functions below - change peak level, change sample rate, change bit rate - may not work with all file formats. Changing peak level and sampling rate always work for .wav files, but that not may be the case for other file formats. Also, except for changing the peak level, the other functions are standalone, that is, they are not run inside the main function of the program
Set the db limit for the audio. For example, to tweak to -5.0 before normalization and to -3.0 after normalization:
audiotame path_to_file -5.0 -3.0
If you wish to tweak the peak only after the normalization:
audiotame path_to_file pass -3.0
audiotame path_to_file sr num
Example:
Change the sampling rate of audio.wav to 44.1khz
audiotame audio.wav sr 44100
audiotame path_to_file convert file_extension
Example:
audiotame audio.wav convert mp3
audiotame path_to_file convert extract
Example:
audiotame vid.mp4 extract
audiotame path_to_file br numk
Example:
Change the sampling rate of audio.wav to 192kbps
audiotame audio.wav br 192k
Check stats of an audio:
audiotame path_to_file stats
acx can be passed as the second argument to verify if an audio complies with ACX's requirements:
audiotame path_to_file acx
| Parameter | Requirement |
|---|---|
| Integrated Loudness | –23 dB to –18 dB |
| Peak Level | ≤ –3 dB |
| Bit Rate | ≥ 192 kbps (CBR) |
| Sample Rate | 44.1 kHz |
This program uses some variables as default:
CONVERT_LOSSY_TO_WAV=1
DB_PEAK_BEFORE_ALL="-100"
DB_PEAK_AFTER_NORM="-100"
NORM_TYPE="ebu"
LOUD_TARGET="-21"
ARNNDN=0
ARNNDN_MODEL="cb.rnnn"
SOX_DENOISE=1
SOX_FACTOR=0.21
SOX_NOISE_THRESHOLD="-50"
SOX_NOISE_MIN_DURATION=0.5
REGULAR_DENOISE=1
REGULAR_NOISE_THRESHOLD="-50"
SILENCE_FLOOR="-60"
DEBUG=0
-
CONVERT_LOSSY_TO_WAV: if the input format is a lossy format, the program will convert to.wav; -
NORM_TYPE: Normalization type:ebuourms; -
LOUD_TARGET: Normalization loudness target; -
ARNNDN: whether to use RNN (ffmpeg's arnndn filter). Must be0or1; -
SOX_DENOISE: whether to denoise (except for arnndn, which is a separete category) the audio or not with sox. Must be0or1. If0; -
SOX_FACTOR: the factor that sox will use to denoise your audio. 0 is none, 1 is maximum. best values are regarded to be in 0.2-0.3 range; -
SOX_NOISE_THRESHOLD: the dB value that sets a threshold below which sound is considered noise by sox; -
SOX_NOISE_MIN_DURATION: the minimum duration that a sound that is below the threshold to be considered noise by sox. The default is 0.5s; -
REGULAR_DENOISE: whether to denoise using ffmpeg's regular filter (lowers intensity of audio below the threshold); -
REGULAR_NOISE_THRESHOLD: he dB value that sets a threshold below which sound is considered noise by ffmpeg; -
SILENCE_FLOOR: the dB value that sets a threshold below which sound is considered silence. -
DEBUG: will print to the console stats of audio files generated at each step.
NORM_TYPE and LOUD_TARGET are set in accordance to ffmpeg-normalize.
Available ARNNDN models are in (https://github.com/richardpl/arnndn-models).
You can override these values with a .env file in the directory from which you run audiotame. Accepted names are audiotame.env, env.audiotame, .env. Or can have a file in ~/.env.audiotame or in ~/.config/.env.audiotame
Example of a .env file:
export NORM_TYPE="rms"
export LOUD_TARGET="-14"
export ARNNDN=1
After installing the package with pip, you can launch gradio with:
audiotame --gradio
Or, you can pull the gradio image
