Multilingual Speech AI

Overview

Multilingual Speech AI is a real-time speech platform supporting live transcription in Darija, French, Arabic and English, natural text-to-speech and a voice assistant that answers simple requests and hands conversations over to a human when needed.

Objectives

• Fine-tune and evaluate speech recognition for Darija.
• Build a low-latency real-time audio pipeline.
• Design the assistant logic and human handover.

Features

• Live transcription in Darija, French, Arabic and English.
• Natural text-to-speech voices.
• Voice activity detection for smooth turn-taking.
• Voice assistant for simple requests.
• Seamless handover to a human agent.

Technologies and profile

faster-whisper, Piper / XTTS, Silero VAD and LiveKit. Profile: AI / ML engineering. Type: Speech AI / Real-time. One to two interns are planned.

Application documents

Attach a CV in PDF format. A short cover letter, GitHub or portfolio link and university internship agreement may also be provided.