Diminish is an implementation of active noise cancellation. It is not generic purpose at this stage though that may eventually be a part of the Roadmap. Right now it is intended as a reference architecture and proof of concept. Please see the documentation for more details.

• Background and Use Case
• Features
• Setup
  • Prerequisites
  • Installation
  • Environment Variables
• Usage
  • Server
  • Client
• Signal Processing Output
• Documentation
• License
• Contributing
• Authors

There are commercial applications that are being considered, but currently this is proof of concept work to experiment with various real world scenarios. See Roadmap for future considerations.

The expected use case for Diminish currently is to provide Active Noise Cancellation (ANC) for a moderate sized room and prevent unwanted signals from entering the hearing of users within said room. The reference microphone is expected to be outside the room by the door and the error microphone, speaker, and ANC processing unit are within the room. Pink Noise is the primary target signal that has been tested with.

• Active Noise Cancellation with Error Microphone, Reference Microphone, Output Speaker
• Reference Microphone over TCP/IP
• Support for multiple algorithms though Recursive Least Squares is the only one currently implemented. See Algorithms for more details
• External C libraries for filtering algorithm and networking - this was necessary to get near real time performance
• Terminal User Interface for server role
• Client User Interface for server and client
• Plotting signals using Matplotlib to enable analysis and debugging

• Python3
• Pip3
• Ubuntu 19 which includes the correct version of PulseAudio. Note version 20 has a higher version that will not work correctly with Diminish.
• gcc compiler
• npm
• docsify

git clone [email protected]:developer-foundry/diminish.git
sudo apt update
sudo apt install build-essential
npm i docsify-cli -g
pip3 install -r requirements.txt
cp environment/.env.example environment/.env

• MODE - options are prerecorded or live. prerecorded performs the active noise cancellation setup on a set of configured files. See INPUT_FILE and TARGET_FILE.
• ALGORITHM - crls is current the only option and is the algorithm that works the fastest and converges the fastest. See algorithm documentation for additional details.
• INPUT_FILE - only needed in prerecorded mode. Defines the input signal. Has only been tested with .wav files.
• TARGET_FILE - required for all modes of operation. The target signal that the application will converge towards. Samples include data/silence.wav and data/pink-noise.wav. Has only been tested with .wav files.
• DEVICE - the hardware device id. Can be modified, but to use your default output and input use DEVICE=default.
• SIZE - only needed in prerecorded mode. Allows user to control the amount of data processed by the algorithm.
• ROLE - specifies is the process is running as SERVER or as CLIENT. SERVER is the primary processing unit and also assumes the speaker and error microphone are attached. CLIENT is a secondary device and assumes the reference microphone is attached.
• WAIT_SIZE - specifices an initialze number of data frames the buffers should hold before processing. A good default value is 5000.
• STEP_SIZE - specifies the number of data frames the algorithm should process in a given run. A good default value is 1024.
• LOGLEVEL - specifies the verbosity of logs generated. The default is INFO.
• SERVER - specifies the IP address (IPv4) of the server. If running as the SERVER role, this can just be set to 127.0.01.
• PORT - specifies the port number the client should connect to the server on. The default should be set to 65432.

Diminish assumes that a client and a server role are running. See architecture documentation for additional details and reference diagrams. The directions below specify the command line tasks to run, but for VSCode users there is also a tasks.json that provides access to these commands via Build tasks.

?> Note it is important that invoke build-libraries is run on both the client and server before the python process is called. This generates the necessary C libraries that are required to execute. See libraries documentation for additional details.

The server can be run using either the terminal user interface (TUI) or the command line interface (CLI). The TUI is useful for running and monitoring and the CLI is useful for debugging.

MODE=live
ALGORITHM=crls
INPUT_FILE=data/truck-and-construction-noises.wav
TARGET_FILE=data/pink-noise.wav
DEVICE=default
SIZE=300000
ROLE=server
WAIT_SIZE=5000
STEP_SIZE=1024
LOGLEVEL=INFO
SERVER=127.0.0.1
PORT=65432

Useful for monitoring the application over time. The TUI specifies keybindings that can be seen on screen to start, pause and quit processing.

python3 -m tui

Useful for debugging or experimentation. The server does support pausing the algorithm processing via a p keybinding.

python3 -m cli

The client can only be run using the command line interface (CLI).

MODE=live
ALGORITHM=crls
INPUT_FILE=data/truck-and-construction-noises.wav
TARGET_FILE=data/pink-noise.wav
DEVICE=default
SIZE=300000
ROLE=client
WAIT_SIZE=5000
STEP_SIZE=1024
LOGLEVEL=INFO
SERVER=192.168.193.224
PORT=65432

Useful for debugging or experimentation.

python3 -m cli

Since the algorithm can be enabled or disabled with p in the CLI or through the TUI, this allows users to see the impact the filtering algorithm has on the output. Once the application is closed, the application will generate a plot of the signals via Matplotlib at plots/{algorithm-name}/{mode}.

In this example it can be seen that the error for the output (red graph) is relatively low and that the error microphone has lower amplitudes in its' wave.

In this example it can be seen that the error for the output (red graph) is high and that the error microphone has higher amplitudes in its' wave.

See the docs folder or navigate to documentation. Documentation is generated via docsify and any terminal videos were generated via asciinema and webgif

Diminish is currently licensed under Apache2

Right now diminish is not accepting contributions or support, but please check back in the future

• Keith LaForce (klaforce)
• Eric LaForce (elaforc)