Spoken Language Identification System

End-to-end spoken language identification for 7 languages: dataset pipeline, MFCC/Mel-spectrogram modelling, error analysis, and a Dockerised Flask deployment

M.Eng. graduation project: a complete audio ML system that identifies which of seven languages (Arabic, English, French, German, Italian, Portuguese, Spanish) is being spoken in an audio clip, delivered as both a mobile and a web application.

This is the full lifecycle, not just a notebook. Dataset construction pulls from Mozilla Common Voice and VoxForge; feature engineering compares MFCC, Mel-spectrogram, and raw-signal (Conv1D) representations; modelling benchmarks classical ML against deep architectures including a VGG-based CNN; and a dedicated error-analysis stage examines where and why the model fails per language.

The Deployment/ folder contains the Flask web application and Dockerfile used to serve inference.

Repository layout

code/    source code, notebooks and datasets
docs/    reports, papers, slide decks and recordings

Contents

Code

• Language_Identification_System/ -- 64 files, 45.4 MB
  • Compare_Lan/
  • CreateDataSet/
  • DataSetCSV/
  • Deployment/
  • Error_Anaylsis/
  • Inferance/
  • MVP/
  • SampleDataSet/
  • .gitignore
  • README.md
  • SLI.png

Documentation

• Language_Identification_System.docx -- 4.5 MB
• Language_Identification_System.pptx -- 8.9 MB

Topics

speech-recognition · audio-classification · deep-learning · mfcc · mel-spectrogram · flask · docker · cnn

---

Part of my AI and Data Science portfolio. Work produced during my M.Eng. in AI and Data Science at the University of Ottawa and my B.Sc. in Computer Science.

Hosam Mahmoud Ibrahim · GitHub · LinkedIn · Portfolio