gogoharrison/multilingual-speech-translator

Real-time speech transcription and multilingual translation API supporting African and international languages. Built with OpenAI Whisper, Helsinki-NLP MarianMT, FastAPI, and Docker. Zero cloud cost — fully local inference. # 🎙️ Multilingual Speech Translation API > Real-time speech transcription and translation pipeline supporting African and international languages — built with OpenAI Whisper, Helsinki-NLP, and FastAPI. --- ## 📌 Overview This project provides a fully containerised, production-ready pipeline that: 1. **Accepts audio input** in any common format (`.wav`, `.mp3`, `.m4a`, `.ogg`, `.flac`) 2. **Transcribes speech to text** using OpenAI Whisper running entirely locally — no API key required 3. **Translates the transcript** using Helsinki-NLP MarianMT models via HuggingFace Transformers 4. **Returns structured JSON** with the transcript, detected language, and translation 5. **Exposes a Streamlit UI** for non-technical users to upload audio and view results instantly The architecture mirrors real-world multilingual conferencing and communication platforms, with specific support for **African languages** (Yoruba, Swahili) alongside French, Spanish, German, and English. > **Zero cloud cost.** All inference runs locally. No OpenAI API key, no Google Cloud billing, no hidden charges. --- ## 🏗️ Architecture ``` ┌─────────────────────────────────────────────────────────────┐ │ Docker Network │ │ │ │ ┌──────────────┐ POST /transcribe ┌───────────────┐ │ │ │ Streamlit │ ──────────────────────▶│ FastAPI │ │ │ │ Frontend │ │ Backend │ │ │ │ :8501 │ ◀──────────────────────│ :8000 │ │ │ └──────────────┘ JSON response └──────┬────────┘ │ │ │ │ │ ┌────────────▼─────────┐ │ │ │ OpenAI Whisper │ │ │ │ (Speech → Text) │ │ │ └────────────┬─────────┘ │ │ …