claverfred/swahili-interpreter

# 🎙️ Real-Time Swahili–English AI Interpretation System **NM-AIST DSAI Capstone Project** **Author:** Fred | ICT & Statistics Unit, AICC | NM-AIST DSAI Program --- ## Overview A cascade neural pipeline that interprets spoken Swahili into English speech in near real-time. ``` Audio (Swahili) ──► ASR ──► Text (Swahili) ──► MT ──► Text (English) ──► TTS ──► Audio (English) Whisper NLLB-200 XTTS-v2 large-v3 distilled-600M ``` --- ## Project Structure ``` swahili-interpreter/ ├── src/ │ ├── asr/ # ASR module (Whisper) │ ├── mt/ # MT module (NLLB-200 + Helsinki-NLP) │ ├── tts/ # TTS module (Coqui XTTS-v2 + Edge-TTS) │ ├── pipeline/ # Full cascade pipeline │ ├── eval/ # WER, BLEU, chrF evaluation │ └── utils/ # Shared utilities ├── configs/ # YAML configuration files ├── notebooks/ # Google Colab notebooks ├── data/ # Audio data (gitignored) ├── results/ # Output files (gitignored) ├── tests/ # Unit tests └── docs/ # Documentation ``` --- ## Quick Start (Google Colab) Open the main notebook directly in Colab: > Replace `YOUR_GITHUB_USERNAME` with your actual GitHub username. --- ## Local Setup (Windows) ```powershell git clone github.com cd swahili-interpreter py -3.10 -m venv venv .\venv\Scripts\Activate.ps1 pip install -r requirements.txt ``` --- ## Pipeline Components | Stage | Model | Device | Metric | |---|---|---|---| | ASR | Whisper large-v3 | GPU float16 | WER | | MT | NLLB-200-distilled-600M | GPU float16 | BLEU / chrF | | TTS | Coqui XTTS-v2 | GPU | RTF / MOS | --- ## Phase 1 Baseline Results | Component | Model | Latency (CPU) | Latency (GPU) | |---|---|---|---| | ASR | Whisper large-v3 | ~103 min | ~3–5 min | | MT | NLLB-200-distilled-600M | 88.05s | ~5–8s | | MT | Helsinki-NLP opus-swc-en | 58 …