README.md

metadata

title: RAFIQ Arabic ASR
emoji: 🎙️
colorFrom: green
colorTo: blue
sdk: docker
pinned: false

RAFIQ – Egyptian Arabic Speech-to-Text API

Custom character-level Transformer trained on Egyptian Arabic audio.
Built as part of the RAFIQ autism-support platform – FCAI, Beni-Suef University 2026.

Usage

1. Upload your model file as entire_model.pth to the Space files.
2. Open /docs for the interactive Swagger UI.
3. POST /transcribe with a WAV/MP3/OGG file (≤ 15 s).

Endpoints

Method	Path	Description
GET	/	API info
GET	/health	Model status + config
POST	/transcribe	Upload audio → Arabic text

Model config

• Character-level vocab (Arabic + English) – 74 tokens
• Encoder: 2 layers, Decoder: 1 layer, d_model=512, 8 heads
• Input: 128-mel spectrogram, 15-second chunks
• Sample rate: 16 000 Hz