Skip to content

Latest commit

 

History

History
67 lines (50 loc) · 2.27 KB

training.md

File metadata and controls

67 lines (50 loc) · 2.27 KB

Speaker Recognition Training Script

This script is used to train a speaker recognition model using audio data. It extracts MFCC features from audio files, preprocesses the data, and trains a neural network model for speaker recognition. The trained model and label encoder classes are saved for later use.

Setup

Prerequisites

  • Python 3.6+
  • NumPy
  • Librosa
  • Scikit-learn
  • TensorFlow

Preprocessing Data

Check helper.md in the helper_functions directory for guidance on preparing data for training.

Installation

  1. Install the required Python packages:

    pip install numpy librosa scikit-learn tensorflow

  2. Organize your audio data in the following directory structure:

    training_data
├── speaker1
│   ├── audio_file1.wav
│   ├── audio_file2.wav
│   └── ...
├── speaker2
│   ├── audio_file1.wav
│   ├── audio_file2.wav
│   └── ...
└── ...

Usage

  1. Update the data_dir variable in the script to point to your training data directory.
  2. Run the script to train the model:
python train_model.py

Script Overview

  • Feature Extraction: MFCC features are extracted from each audio file using the extract_features function.
  • Data Preparation: Features and corresponding speaker labels are collected from all audio files.
  • Label Encoding: Speaker labels are encoded using LabelEncoder from scikit-learn.
  • Train-Test Split: Data is split into training and testing sets using train_test_split from scikit-learn.
  • Model Architecture: The neural network model architecture is defined using Sequential from Keras.
  • Model Compilation: The model is compiled with the Adam optimizer and sparse categorical cross-entropy loss.
  • Model Training: The model is trained on the training data for 50 epochs with a batch size of 32.
  • Model Saving: The trained model and label encoder classes are saved to disk for later use.

File Structure

  • train_model.py: The main training script.
  • training_data: Directory containing audio files organized by speaker.

Notes

Ensure the audio files are in WAV format and have a sample rate of 16,000 Hz. This script assumes each speaker has a maximum of 1800 audio files. Adjust the loop logic if your data differs.