Speechmatics Debuts World-First Arabic-English Medical Voice AI

CAMBRIDGE, England – March 10, 2026 – Voice technology firm Speechmatics today announced a significant advancement in multilingual artificial intelligence with the launch of its Arabic–English bilingual model. The system is engineered to seamlessly understand and transcribe conversations that switch between Arabic dialects and English, a common linguistic pattern in the Middle East and North Africa (MENA). Accompanying this release is a specialized clinical variant, which the company bills as the world's first Arabic–English bilingual medical model, designed to bring new levels of accuracy to healthcare documentation.

This development addresses a persistent challenge for standard speech-to-text systems, which often falter when speakers blend languages mid-sentence. By creating a single, production-ready model that handles both languages simultaneously, Speechmatics aims to provide a more reliable solution for critical sectors like healthcare, finance, and customer service across the region.

Cracking the Code-Switching Challenge

In professional settings throughout the MENA region, from a Riyadh contact center to a Cairo hospital, conversations naturally flow between Arabic and English. This phenomenon, known as code-switching, has long been a stumbling block for monolingual AI models. When a speaker shifts languages, traditional systems can lose context, leading to misattributed words, dropped terminology, and inaccurate transcripts. In high-stakes environments, such errors are not minor inconveniences but significant operational risks.

Speechmatics' new model is built specifically to overcome this hurdle. According to the company, it delivers a 35% lower Word Error Rate (WER) than Google on complex Arabic–English code-switching tasks, achieving an error rate of 6.3% compared to Google's 9.7%. The lower WER indicates a substantially more accurate transcription. The model also excels in monolingual performance, outperforming major providers on Arabic-only transcription with a 24% lower WER than Google (4.5% vs. 5.9%) and surpassing the accuracy of systems from OpenAI Whisper, AssemblyAI, Deepgram, Amazon, and Microsoft.

Beyond just language, the system is trained to recognize and differentiate between various Arabic dialects, including Gulf, Egyptian, and Levantine variations. This is a crucial distinction, as models trained solely on Modern Standard Arabic—the formal version often used in broadcasting—struggle to comprehend the nuances of everyday conversational speech. The new model includes features like speaker diarization and speaker focus as standard, ensuring that every word is correctly attributed to the right person, even in multi-participant discussions.

A New Frontier for Medical AI in the Middle East

The most groundbreaking component of the launch is the specialized bilingual medical model. In clinical settings across MENA, it is routine for healthcare professionals to discuss patient care using a mix of Arabic and English, with drug names, procedural terms, and dosages often stated in English. Generic transcription models frequently mishandle this specialized vocabulary, introducing errors that can compromise patient records and care.

Trained on a vocabulary twice the size of the company's existing English Medical Model, this new variant incorporates extensive English and Arabic clinical terminology, real-world dialect variations, and audio from actual clinical settings. It is designed to accurately transcribe complex information such as ICD-10-CM diagnostic codes, pharmaceutical names, and clinical shorthand, regardless of which language is used.

The immediate impact of this technology is being validated through partnerships. Sully.ai, a company focused on scaling healthcare AI infrastructure, evaluated the model for its regional expansion. "We ran extensive evaluations on complex clinical audio, including code-switching and dialect-heavy consultations common across MENA," said Patrick Nguyen, Head of Engineering, MENA, at Sully.ai. "Speechmatics' bilingual medical model was the only one that met the performance thresholds we require to maintain high-quality clinical documentation as we scale regionally. That alignment made the partnership a strong fit for our expansion."

By ensuring the fidelity of clinical notes, the technology promises to streamline workflows, reduce the administrative burden on clinicians, and ultimately enhance patient safety.

Addressing Data Sovereignty with Enterprise-Ready Deployment

A key barrier to AI adoption in the MENA region has been the strict data sovereignty and privacy laws. Nations like Saudi Arabia, with its Personal Data Protection Law (PDPL), and the United Arab Emirates have established robust legal frameworks that dictate where and how sensitive data, including voice data, can be processed and stored. For regulated industries such as healthcare and finance, these laws often mandate that data remain within national borders.

Speechmatics has directly addressed this requirement by offering flexible deployment options. The new models can be deployed via a cloud SaaS platform, on-premises within an organization's own data centers, or even directly on-device. This versatility allows enterprises to maintain full control over their data, ensuring compliance with local regulations and alleviating concerns about cross-border data transfers. This capability is critical for building trust and enabling the adoption of AI in privacy-critical applications.

The system's architecture, powered by NVIDIA AI infrastructure and optimized for high-throughput processing, maintains sub-second latency for real-time transcription across all deployment modes. This ensures that performance is not sacrificed for security, providing a seamless experience for both real-time streaming and batch transcription workflows.

The Broader Impact on Global Communication

The launch represents more than just a product update; it signifies a maturing of voice AI technology to better reflect the linguistic reality of a globalized world. The global voice and speech recognition market is projected to grow to over $53 billion by 2030, with healthcare being a primary driver. By tackling the difficult problem of code-switching head-on, Speechmatics is setting a new benchmark for what businesses and users should expect from voice AI.

“This was critical to achieving meaningful outcomes for customers across the region who kept describing the same challenge,” explained Katy Wigdahl, CEO of Speechmatics. “In a Cairo hospital or a Riyadh contact center, Arabic and English flow concurrently - the drug name arrives in English, the rest of the sentence is Arabic. Delivering significant impact meant removing that friction from voice interactions. We trained on real voices, real dialects and real clinical vocabulary - because that’s the only way to build something that truly works where it’s used.”

As AI becomes more integrated into daily professional life, the ability to understand and operate within multilingual contexts will be paramount. With this launch, Speechmatics has not only provided a powerful tool for the MENA region but has also demonstrated a path forward for developing more inclusive and effective AI for a world that rarely speaks in a single language. Both the general and medical bilingual models are available now for enterprise deployment.