themrandroid/native-language-accent-classifier-using-knn

Predict Hausa, Igbo, or Yoruba from English accent audio # Native Language Accent Classifier (Hausa, Igbo, Yoruba) A compact, reproducible pipeline that classifies a speaker's native language (Hausa, Igbo, Yoruba) from short audio samples using engineered audio features (MFCCs, pitch, ZCR, spectral contrast) and a K-Nearest Neighbors (KNN) model. The repository contains audio collection utilities, preprocessing, feature extraction, training/evaluation code in a notebook, saved feature CSVs, and a Streamlit demo app. Key goals - Reliable feature extraction from short audio samples. - Reproducible preprocessing and augmentation / bootstrapping. - Simple, explainable baseline model for rapid iteration and demo. Primary code symbols (open these for details) - `app.extract_features`: feature extraction used in the Streamlit app. See app.py. - `native_language_accent.preprocess_audio`: audio conversion, normalization, trimming routine. See native_language_accent.ipynb. - `native_language_accent.extract_features`: dataset feature extraction loop (13 MFCC means, pitch mean/std, zcr, 7 spectral contrast means). See native_language_accent.ipynb. - `native_language_accent.download_audio`: YouTube/shorts download helper used to populate data/. See native_language_accent.ipynb. Quick start 1) Prerequisites - Python 3.8+ (tested). - System dependency: ffmpeg (required by pydub). - Recommended: create a virtual environment. 2) Install Python dependencies Run: ```bash python -m pip install -U pip python -m pip install numpy pandas librosa scikit-learn matplotlib seaborn pydub yt_dlp streamlit joblib ``` Or: ```bash python -m pip install -r requirements.txt ``` (If you add a requirements file, put it at repo root.) 3) Reproduce feature extraction and training (recommended) - Open and run native_language_accent.ipynb. - Use `native_language_accent.preprocess_audio` to convert files from data/ into processed_data/. - Use `native_language_accent.extract_features` to build `accent_features.csv`. …