CNTXT AI announced the launch of Munsit Edge, one of the first Arabic speech-to-text systems to run fully on-device, with no cloud connection required.
Built on the same Munsit Arabic ASR (automatic speech recognition) model that has set a global benchmark for transcription accuracy across major Arabic dialects, Munsit Edge delivers real-time Arabic transcription on consumer hardware.
For the first time, the Arabic speech-to-text model runs entirely on the device, where the conversation happens: on a phone in Riyadh, in a car in Cairo, in a home in Dubai, in a contact centre in Manama. Nothing leaves the user’s hands.
“Until today, the Arab world has never had Arabic speech recognition that truly ran on the devices people use,” said Mohammad Abu Sheikh, CEO of CNTXT AI. “With Munsit, we proved you can build an Arabic‑first model that beats generic systems on accuracy. With Munsit Edge, we’ve moved that model out of distant data centres and onto the devices themselves. Your calls, your cars, your homes – all of them can now understand Arabic in real time without sending a single second of audio to a third‑party cloud.”
A first for Arabic speech recognition
Until now, all Arabic STT systems have depended on cloud inference. Audio is streamed to remote servers, processed on GPUs, and returned across the network, introducing latency, privacy exposure, and infrastructure cost at every interaction. The linguistic complexity of Arabic (rich dialect diversity, frequent code-switching with English, demanding real-world acoustic conditions) has made on-device deployment particularly difficult.
Munsit Edge solves it. Through a foundation-trained Arabic ASR model and runtime optimisation tuned for everyday consumer hardware, Munsit Edge runs across iPhone, Android, MacBook, Windows and Linux PCs, in-car systems, smart home devices, and embedded IoT, at speeds and accuracy levels that, until today, required server infrastructure.
Performance highlights:
- Around 24% word error rate (WER – accuracy metric in speech recognition), across major Arabic dialects: Gulf, Egyptian, Levantine, MSA, and Arabic-English code-switching
- Latency around 150 ms for real‑time streaming transcription on a standard iPhone‑class device
- Same Arabic accuracy across cloud, on‑prem, and on‑device deployments. No quality trade-off for moving on-device.
- No network connection is required for inference. Transcription runs fully locally.
Why on-device matters
“Sovereignty isn’t just where your data is stored. It’s where it’s processed,” Abu Sheikh added. “For a bank handling a customer’s voice or a government running a citizen helpline, what matters is that Arabic speech can be understood on their own infrastructure. Munsit Edge makes it possible to do that on phones, PCs, and edge devices across the region.”
Munsit Edge is purpose-built for the use cases where on-device matters most:
- Contact centres, telco and IVR. Real-time Arabic call transcription with zero per-minute server cost, deployed inside the operator’s own infrastructure.
- Banking and fintech. Voice transcription that meets strict data-residency requirements.
- Government and public sector. Citizen-facing voice services on sovereign infrastructure.
- In-car and embedded systems. Arabic voice interfaces that work without a cellular connection.
- Smart home and consumer devices. Arabic voice control without a cloud subscription, working offline.
Availability
Munsit Edge is available today through three integration paths:
- Native SDKs for iOS, Android, macOS, Windows, and Linux
- On-premise containers for private cloud and data centre deployment
- Embedded IoT builds for automotive, smart home, and industrial hardware
The broader Munsit platform also remains available via secure cloud API, web workspace, and mobile app, so organisations can combine cloud and on‑device deployments based on their data and latency needs.
Discussion about this post