audx

A flexible audio processing tool for transcoding, noise reduction, and filtering. Features a modular processor-based pipeline for professional audio workflows.

Demo

Listen to the noise reduction in action:

Before (Noisy)	After (Clean)
▶️ Play audio	▶️ Play audio

Features

• Noise Reduction: RNNoise integration with embedded model and Voice Activity Detection, plus support for custom models
• Universal Format Support: Decode/encode MP3, AAC, Opus, FLAC, ALAC, WAV, and any FFmpeg-supported format
• Audio Filtering: Tempo, volume, EQ, normalization, effects, and full FFmpeg filter chain support
• Quality Control: Explicit bitrate control for all codecs
• Modular Architecture: Extensible processor pipeline for custom audio processing workflows
• Automatic Conversion: Sample rate and channel layout conversion handled transparently

Installation

Dependencies

Component	Version	Description
CMake	4.0+	Build system
GCC/Clang	C99+	Compiler
FFmpeg	8.0+	Audio codec libraries
RNNoise	0.2+	Noise reduction library

Quick Build

Debian/Ubuntu:

sudo apt install cmake build-essential libavcodec-dev libavformat-dev \
  libavutil-dev libavfilter-dev libswresample-dev libswscale-dev
./scripts/install-rnnoise.sh
./scripts/build.sh

Arch Linux:

sudo pacman -S cmake gcc ffmpeg
./scripts/install-rnnoise.sh
./scripts/build.sh

Output: build/bin/audx

Usage

audx <input> <output> [OPTIONS]

Options

Option	Short	Description	Example
--help	-h	Show help message
--version	-v	Show version information
--codec=<name>	-c	Output codec (auto-detects from extension if not specified)	--codec=libopus
--bitrate=<rate>	-b	Bitrate (e.g., 192k, 320k)	--bitrate=192k
--sample-rate=<hz>	-r	Output sample rate in Hz	--sample-rate=48000
--channels=<n>		Output channels (1=mono, 2=stereo, up to 8)	--channels=2
--filter=<expr>	-f	FFmpeg filter expression	--filter="atempo=1.25,volume=0.5"
--denoise	-d	Enable RNNoise denoising (uses embedded model)	-d
--model=<path>		Custom RNNoise model file (requires -d/--denoise)	--model=/path/to/model.bin
--denoise-vad-threshold=<n>		Voice Activity Detection threshold (0.0-1.0, default: 0.5)	--denoise-vad-threshold=0.7
--list-models		List available RNNoise models

Supported Codecs

Lossy Codecs

Codec	Name	Quality Bitrates
MP3	libmp3lame	128k / 192k / 256k / 320k
AAC	aac	96k / 160k / 256k / 320k
Opus	libopus	96k / 128k / 192k / 256k

Lossless Codecs

Codec	Name	Compression Levels
FLAC	flac	5 / 8 / 10 / 12
ALAC	alac	5 / 8 / 10 / 12

Raw Audio

Format	Name	Description
WAV/PCM	pcm_s16le	16-bit signed little-endian PCM

Examples

Noise Reduction

Remove background noise from voice recording:

audx noisy_audio.wav clean.mp3 -d --codec=libmp3lame --bitrate=256k

Aggressive noise removal for podcast:

audx podcast.wav output.mp3 -d --denoise-vad-threshold=0.7 \
  --codec=libmp3lame --bitrate=192k

Denoise with custom model:

audx input.wav output.wav -d --model=/path/to/custom_model.bin

Denoise and normalize:

audx interview.wav clean.mp3 -d \
  --filter="highpass=f=80,loudnorm=I=-16:TP=-1.5" \
  --codec=libmp3lame --bitrate=256k

Format Conversion

MP3 to Opus:

audx input.mp3 output.opus --codec=libopus --bitrate=192k

FLAC to high-quality MP3:

audx input.flac output.mp3 --codec=libmp3lame --bitrate=320k

Convert to lossless FLAC:

audx input.mp3 output.flac --codec=flac

Audio Processing

Speed up audio by 25%:

audx input.mp3 output.mp3 --codec=libmp3lame --bitrate=256k --filter="atempo=1.25"

Reduce volume by 50%:

audx input.mp3 output.mp3 --codec=libmp3lame --bitrate=256k --filter="volume=0.5"

Chain multiple filters:

audx input.mp3 output.mp3 --codec=libmp3lame --bitrate=320k \
  --filter="atempo=1.25,volume=0.8,highpass=f=100"

Loudness normalization:

audx input.mp3 output.mp3 --codec=libmp3lame --bitrate=256k \
  --filter="loudnorm=I=-16:LRA=11:TP=-1.5"

Common Filters

Filter	Purpose	Example
atempo	Change playback speed	atempo=1.5 (50% faster)
volume	Adjust volume	volume=0.5 (50% quieter)
loudnorm	Loudness normalization	loudnorm=I=-16:TP=-1.5:LRA=11
highpass	High-pass filter (cut low freq)	highpass=f=200 (cut below 200Hz)
lowpass	Low-pass filter (cut high freq)	lowpass=f=3000 (cut above 3kHz)
aecho	Echo effect	aecho=0.8:0.88:60:0.4
equalizer	Parametric EQ	equalizer=f=1000:width=200:g=-10
aresample	Resample audio	aresample=48000

Full documentation: https://ffmpeg.org/ffmpeg-filters.html#Audio-Filters

Technical Notes

Opus Sample Rates

Opus only supports: 48000, 24000, 16000, 12000, 8000 Hz

For other sample rates, add resampling filter:

audx input.mp3 output.opus --codec=libopus --quality=high --filter="aresample=48000"

RNNoise Processing

• Default model: Uses embedded RNNoise model (no external files needed)
• Custom models: Support for custom RNNoise models via --model flag
• Internal processing: 48kHz mono (automatic resampling)
• Frame size: 480 samples (handled automatically via buffering)
• Multi-channel: Each channel processed independently
• VAD threshold: 0.5 default; lower = more aggressive, higher = more conservative

Default Behavior

Without --codec, audx outputs raw PCM data for backward compatibility.

Architecture

audx uses a modular processor-based pipeline architecture:

Input File → Decoder → [Denoiser] → [Filter] → Encoder → Output File
             AVFrame    AVFrame       AVFrame    AVFrame

Processors

Processor	Location	Purpose
Decoder	src/core/processors/decoder.c	FFmpeg decoding + format conversion
Denoiser	src/core/processors/denoiser.c	RNNoise with VAD support
Filter	src/core/processors/ffmpeg_filter.c	FFmpeg filtergraph integration
Encoder	src/core/processors/encoder.c	FFmpeg encoding with quality presets

Key Features

• Plugin Architecture: Easy to extend with custom processors
• Format Negotiation: Automatic conversion between processor requirements
• Frame Buffering: Handles fixed-size requirements (RNNoise 480 samples, codec frame sizes)
• Pipeline Orchestration: Manages processor lifecycle and error handling

For developers: See REDESIGN.md for full architecture documentation and CLAUDE.md for development guidelines.

License

audx source code is licensed under the MIT License.

FFmpeg libraries used by audx are licensed under the LGPL 2.1 or later.

Important Legal Information

For Commercial Use: Some audio codecs (particularly AAC) may require patent licenses in certain jurisdictions, independent of copyright licensing. Consult with a lawyer before using audx in commercial products.

Complete licensing information, FFmpeg compliance details, and patent notices: See LEGAL_NOTICES.md

Quick Compliance Summary

• Personal/Non-Commercial Use: Free to use with all codecs
• Commercial Use: Verify patent requirements for your jurisdiction
• Distribution: Must include LICENSE and LEGAL_NOTICES.md files
• Modifications: Must maintain MIT license for audx, LGPL for FFmpeg libraries

For FFmpeg legal information: https://www.ffmpeg.org/legal.html

References

• FFmpeg Documentation: https://ffmpeg.org/documentation.html
• FFmpeg Audio Filters: https://ffmpeg.org/ffmpeg-filters.html#Audio-Filters
• FFmpeg Codecs: https://ffmpeg.org/ffmpeg-codecs.html