/Capuchinbird-Call-Detector

Primary LanguageJupyter Notebook

Capuchinbird Call Detector

This project aims to build an audio classifier using TensorFlow to detect Capuchinbird calls from audio recordings. The key steps include:

  1. Installation: Installing necessary libraries like TensorFlow, TensorFlow-IO, and Matplotlib.
  2. Data Preparation: Loading audio files, preprocessing them by resampling to 16kHz, and converting them to spectrograms.
  3. Model Training: Creating and training a Convolutional Neural Network (CNN) on labeled audio data of Capuchinbird calls and other sounds.
  4. Prediction on New Data: Applying the trained model to new audio recordings to detect Capuchinbird calls, using a sliding window approach to handle long recordings.
  5. Export Results: Aggregating the model predictions and exporting the results to a CSV file.

Table of Contents

  1. Installation
  2. Data Preparation
  3. Model Training
  4. Prediction on New Data
  5. Export Results
  6. Credits
  7. License

Installation

Install required libraries:

pip install tensorflow==2.8.0 tensorflow-io==0.25.0 matplotlib

Data Preparation

Load and preprocess audio files, converting them to a standard format suitable for model training.

Model Training

Train a CNN on the preprocessed audio data to classify Capuchinbird calls.

Prediction on New Data

Use the trained model to detect Capuchinbird calls in new audio recordings and post-process the results to identify distinct calls.

Export Results

Export the final detection results to a CSV file for analysis.

Credits

This project is based on the tutorial by Nick Nochnack. For more information, visit Nick Nochnack's GitHub.