This repository contains code and instructions for reproducing the research described in the paper Font, F., & Serra, X. (2016). Tempo Estimation for Music Loops and a Simple Confidence Measure. In Int. Conf. on Music Information Retrieval (ISMIR). The full text of the paper can be found here.
In order to run the experiments described in the paper you'll need to set up the
datasets and analyze its content. You should create a Python virtual environment and install the
requirements listed in requirements.txt. In addition, you'll need to
install ffmpeg (for audio conversion) and, optionally,
rabbitMQ (needed for paralelizing analysis using Celery
distributed task manager). Then you should follow instructions below:
- Instructions for setting up datasets
- Instructions for setting up analysis algorithms
- Instructions for analyzing the datasets
Once datasets are set up and audio analysis has been carried out, you can open the following IPython notebooks which contain the code to generate the results and plots shown in the paper:
- Datasets: information and statistics about the datasets, corresponds to Section 4.1 of the paper.
- Confidence measure: description of the confidence measure with examples and code, corresponds to Section 3 of the paper.
- Tempo estimation results: evaluation of the different tempo estimation algorithms and confidence measure, corresponds to Section 5 of the paper.
UPDATE: we implemented Percival's BPM estimation method in Essentia (see PercivalBpmEstimator algorithm). The following notebooks compare the results of the Essentia implementation and the original Python implementation provided by the authors (notebook1, notebook2).