Featured project

ARconvo

Independent build · Product builder and developer

Live / Working

Bilateral Arabic–English speech-translation interface for two-way spoken communication that runs fully offline, with no dependence on internet connectivity.

Discuss this work

Purpose

ARconvo supports two-way Arabic-English spoken communication entirely offline, making it practical where low connectivity or privacy-sensitive environments rule out cloud inference.

Highlights

Automatic language detection and silence detection
Faster-Whisper speech recognition with an Arabic-dialect-trained model
Falcon 3B translation with native and model-based speech synthesis
Optimised to run on a PC with 4GB VRAM

Capabilities

Fully offline local-PC operation
Automatic language and silence detection
Faster-Whisper speech recognition
Arabic-dialect-trained speech workflow
Falcon 3B translation
Windows-native English TTS and model-based Arabic voice synthesis
Arabic and English transcript capture

Governance & Quality

Iterative evaluation with native Arabic-speaking testers
Validation against Arabic detection, translation accuracy and interaction quality

Architecture & Delivery

Optimized for a PC with 4 GB VRAM
Designed around local inference and offline edge deployment constraints

Outcomes

Working local application for bilateral offline speech translation
Demonstrates practical voice AI under constrained hardware conditions

Skills Demonstrated

Conversational AI
Voice AI
Multilingual NLP
Arabic dialect handling
Speech recognition
Voice synthesis
Translation
Edge AI
Offline inference
Model optimisation