Tools
librosa
librosa is an open-source Python library designed for music and audio analysis. It provides functionalities for handling audio signals in a variety of formats such as .wav, .mp3, and .flac. These building blocks enable the reading and writing of audio files, computing of spectrograms, extracting of features like tempo and pitch, and visualizing of data. librosa is widely used for audio signal processing and machine learning applications involving audio data [1].
​
In this project, librosa was instrumental in facilitating audio source separation by performing NMF and HPSS on regular spectrograms and Mel spectrograms. These techniques are further described on the Techniques page of this website.​​
MATLAB
MATLAB is a high-level programming language and interactive environment developed by MathWorks, used by engineers, scientists, and researchers. It is popular for signal processing, control systems, image processing, and machine learning. We used MATLAB to do matrix/vector calculations of audio files such as NMF. ​
Spleeter
Spleeter is an open-source deep learning-based music source separation Python library developed by Deezer, a music streaming platform. It efficiently separates audio tracks into individual stems using Tensorflow and includes pre-trained models capable of performing several source separations:
​
-
Two stems: Vocals / accompaniment separation
-
Four stems: Vocals / drums / bass / other separation
-
Five stems: Vocals / drums / bass / piano / other separation​
​
In this project, Spleeter was utilized to separate the mixed audio file into its bass and drum components. The separation results are quantitatively evaluated against the original bass and drum tracks to provide a benchmark for ideal source separation, as Spleeter is widely recognized and used in the field of music source separation.​
mir_eval
mir_eval is an open-source Python library designed to evaluate source separation results. It includes a collection of standardized tools that ensure consistent assessment across different techniques and applications.
​
In this project, mir_eval was used to implement standard measures such as SDR, SIR, and SAR. These metrics are further described on the Evaluation page of this website.