Gender Recognition By Voice

A collection of scripts for training and evaluating machine-learning models that classify a speaker's gender from short voice recordings. The project includes utilities for data preprocessing, model comparison, neural network training, and collecting new audio samples for inference.

Dataset

• voice.csv – tabular dataset containing acoustic features for labeled male and female voice samples. The last column stores the target label; all other columns are features that are scaled before training.
• output/voiceDetails.csv – generated dynamically by the R script when you analyse a freshly recorded voice sample. The file is preprocessed to match the format of voice.csv before inference.

Requirements

Python

• Python 3.7+
• scikit-learn
• matplotlib
• pandas
• numpy
• pyaudio (required only when recording new samples)

Install the Python dependencies from the included requirements.txt file:

pip install -r requirements.txt

If you do not plan to use the recording workflow, you may omit installing pyaudio.

R

• R 3.5+
• warbleR

Install the R dependency from the R console:

install.packages("warbleR")

Usage

1. Compare multiple classifiers

Run the comparison script to evaluate several classical models (k-NN, SVM, Decision Tree, Random Forest, MLP) on the dataset:

python clf_comparison.py

The script prints training and testing metrics for each classifier.

2. Train and evaluate the neural network

Train an MLPClassifier on voice.csv and report detailed metrics:

python neural_net.py

After training, the model is saved as trained_neural_net and the loss curve is displayed using Matplotlib.

3. Interactive menu for end-to-end flow

The main entry point exposes a small CLI menu that lets you train the neural network or record a new voice sample for inference:

python main.py

Menu options:

1. Train Neural Net – runs the same training routine as neural_net.py and persists the model to trained_neural_net.
2. Analyse Voice – records a 20-second sample (requires a microphone and pyaudio), extracts features via the R script getAttributes.r, preprocesses them to match the dataset, and prints the predicted gender.
3. Exit – closes the menu.

The recorded audio is stored at sounds/output.wav, and the generated features are saved inside the output/ directory. Ensure that trained_neural_net exists before choosing the Analyse Voice option; otherwise, train the model first.

Project Structure

• data_process.py – shared routines for loading, scaling, visualising, and splitting the dataset, as well as computing evaluation metrics.
• clf_comparison.py – trains and evaluates a suite of traditional classifiers.
• neural_net.py – trains, evaluates, and persists the neural network classifier.
• main.py – CLI interface that wires together training, recording, and inference.
• sound_recorder.py – handles audio capture and saving the recorded waveform.
• getAttributes.r – extracts acoustic features from recorded audio using warbleR.

Troubleshooting

• Missing audio device: pyaudio requires access to an input device. On headless systems, you may need to configure a virtual microphone or skip the recording workflow.
• R script errors: Ensure that R is installed and available in your system PATH, and that the warbleR package is installed. The CLI invokes the script using Rscript getAttributes.r from the project root.
• Matplotlib backend issues: If plots fail to display on headless servers, switch to a non-interactive backend by exporting MPLBACKEND=Agg before running the scripts.

License

This repository does not include an explicit license. Please contact the original authors before using the code in production.