Logo Lanfrica

praiseikebuwa/African-Speech-AI

Domain:

natural language processing

Record type:

software
Creator:
pra
Host:
A complete, production-ready Python-based multilingual speech AI system supporting English, Yoruba, Igbo, Hausa, and Nigerian Pidgin. Built with PyTorch, FastAPI, and modern deep learning practices. # πŸŽ™οΈ African Speech AI A complete, production-ready Python-based multilingual speech AI system supporting **English, Yoruba, Igbo, Hausa, and Nigerian Pidgin**. Built with PyTorch, FastAPI, and modern deep learning practices. --- ## πŸ“ Full Project Structure ``` african_speech_ai/ β”œβ”€β”€ main.py # Unified CLI entry point β”œβ”€β”€ setup.py # Package installer β”œβ”€β”€ requirements.txt # All dependencies β”‚ β”œβ”€β”€ config/ β”‚ β”œβ”€β”€ __init__.py β”‚ β”œβ”€β”€ settings.py # All hyperparameters & paths (single source of truth) β”‚ └── languages.py # Vocab, tone maps, slang, intent/entity labels β”‚ β”œβ”€β”€ audio/ β”‚ β”œβ”€β”€ __init__.py β”‚ β”œβ”€β”€ capture.py # Real-time mic capture + streaming utterance detection β”‚ β”œβ”€β”€ vad.py # WebRTC Voice Activity Detection + energy fallback β”‚ β”œβ”€β”€ preprocessing.py # Noise reduction, pre-emphasis, normalisation β”‚ └── features.py # Mel-spectrogram / MFCC extraction with delta features β”‚ β”œβ”€β”€ asr/ β”‚ β”œβ”€β”€ __init__.py β”‚ β”œβ”€β”€ model.py # CNN + BiLSTM / Transformer ASR model with CTC β”‚ β”œβ”€β”€ tokenizer.py # Character-level multilingual tokenizer β”‚ β”œβ”€β”€ decoder.py # Greedy + beam-search CTC decoders β”‚ └── inference.py # Full ASR pipeline (file/bytes/stream β†’ text) β”‚ β”œβ”€β”€ languages/ β”‚ β”œβ”€β”€ __init__.py β”‚ β”œβ”€β”€ yoruba.py # Tonal processing, syllabification, IPA, slang β”‚ β”œβ”€β”€ igbo.py # Igbo processing, tone, IPA β”‚ β”œβ”€β”€ hausa.py # Hausa ejectives, tone, IPA β”‚ └── normalizer.py # Language detection, unified normaliser, transliterator β”‚ β”œβ”€β”€ nlp/ β”‚ β”œβ”€β”€ __init__.py β”‚ β”œβ”€β”€ intent.py # Transformer intent classifier + keyword fallback β”‚ β”œβ”€β”€ entities.py # Transformer NER + rule-based NER (phone/money/URL) β”‚ β”œβ”€β”€ context.py # Multi-turn conversation context + session …