The repository provides sample apps for all desktop platforms demonstrating Krisp SDK functionality. The build system and the codebase is compatible with.
- Mac M1
- Mac Intel
- Windows(amd64)
- Linux(amd64)
The reference samples require
- libsndfile library to read and write WAV files
- Krisp SDK package with archive libraries for the noise canceling
The following environment variables are mandatory. The name of each parameter is self-explanatory.
- KRISP_SDK_PATH
- LIBSNDFILE_INC
- LIBSNDFILE_LIB
KRISP_SDK_PATH should point to the Krisp SDK package directory.
LIBSNDFILE_INC and LIBSNDFILE_LIB directories is required by the sample-nc app for the purpose of reading and writing WAV PCM-based audio files. It is not the SDK requirement.
brew install libsndfile
In addition to above dependencies you will need the pybind11 and NumPy library to build CPython modules using C++.
sudo apt-get install pybind11-dev
pip3 install numpy
brew install pybind11
pip3 install numpy
How to run the build
For Krisp NC SDK
make
For Krisp AL SDK
make al
make vs
Node v20 or above. NAPI Version 9. In addition to above depencies you will need the npm, and deps defined in the src/sample-node/package.json.
The NODE_INC environement variable should be set to the include directory of the installed Node.
On ARM based Mac with Homebrew the path could be /opt/homebrew/include/node
On Ubuntu Linux with nvm it could be the $HOME/.nvm/versions/node/v22.9.0/include/node
if installed locally.
make node
make python
Run build-vs-solution.bat. Open the Visual Studio Solution located in the vs-solution folder and manually build the target apps.
This will produce a Visual Studio Solution project in the vsbuild folder.
All apps will be stored inside the bin folder in the root directory
The noise cancelling app that applies Krisp NC technology on the given PCM16 wav file using given Krisp Weight file model. The app with its codebase demonstrates
- how to initialize Krisp SDK and how to free memory resources if you don't need to use Krisp anymore
- how to load a single model
- how to define the size of the frame to prepare the SDK for the processing of the frame sequence
- how to process the sound frame-by-frame using Krisp
- how to process either PCM16 or FLOAT based audio file
- how to use either regular Krisp NC API or NC API with Call Stats
- how to to get the Call Stats for the whole processed file and for the each frame, the feature is enabled with '-s' option
sample-nc -i <PCM16 or FLOAT32 wav file> -o <output WAV file path> -m <path to the AI model> -s
The sample imitates realtime PCM16 audio stream by reading PCM16 WAV file. It uses CPython based wrapper over Krisp Audio SDK to process audio data. The processed output is stored in the WAV file.
python3 process_wav.py -i <PCM16 wav file> -o <output WAV file path> -m <path to the AI model>
The sample imitates realtime PCM16 or FLOAT32 audio stream by reading PCM16/FLOAT32 WAV file. It uses Node based wrapper over Krisp Audio SDK to process audio data. The processed output is stored in the WAV file.
cd src/sample-node
node index.js -i <PCM16 wav file> -o <output WAV file path> -m <path to the AI model>
The sample demonstrates how to build dynamic link library using Krisp static libraries.
Dynamic link libraries for Linux are always bound to specific GLIBC version. It means that you will not be able to use them on older Linux systems. This samples can be used to build DLL libraries for old Linux systems using Krisp static libraries.