Speechmatics has launched an Arabic-English bilingual model, a single production-ready system that processes Arabic dialects and English simultaneously, the Cambridge-based company said on March 10, 2026. The model achieves a 6.3% Word Error Rate on mixed-speech benchmarks, outperforming the nearest competitor by 35%.
The release also introduces the first Arabic-English bilingual medical model in the world. This clinical variant is trained on twice the vocabulary of Speechmatics’ existing English Medical Model. It incorporates English and Arabic clinical terminology, real dialect variation, and speech recorded in actual clinical settings.
Addressing Code-Switching in MENA
Across the Middle East and North Africa, speakers routinely shift between Arabic and English within a single sentence. This pattern, known as code-switching, causes standard monolingual models to misattribute words, drop terminology, or produce incorrect transcriptions. In contact centers and clinical environments, such errors are common rather than exceptional.
The Arabic-English bilingual model is built to handle this directly. In benchmarking tests, Speechmatics recorded a 35% lower Word Error Rate than Google on code-switching tasks, achieving 6.3% compared to Google’s 9.7%. The model also includes speaker diarization and speaker focus, ensuring each word is attributed to the correct speaker throughout a session.
Dialect Coverage Across the Region
Arabic varies significantly across the region, with Gulf, Egyptian, and Levantine dialects each carrying distinct vocabulary, phonology, and rhythm. Models trained on Modern Standard Arabic alone perform poorly in real conversational settings.
On Arabic-only transcription, Speechmatics achieves a 24% lower Word Error Rate than Google, recording 4.5% against Google’s 5.9%. The company said its model also outperforms OpenAI Whisper, AssemblyAI, Deepgram, Amazon, and Microsoft on this metric.
Enterprise Deployment and Data Sovereignty
Data sovereignty requirements across MENA, including regulations in Saudi Arabia and the UAE, place strict obligations on where voice data is processed and stored. The model supports deployment across cloud SaaS, on-premises, and on-device environments. It runs on NVIDIA AI infrastructure and is optimized through NVIDIA Dynamo-Triton for high-throughput, low-latency processing. Sub-second latency is maintained across all deployment modes.
Real-time streaming and batch transcription operate on the same model, removing the accuracy trade-off that typically accompanies switching between the two modes. Punctuated transcripts, timestamped outputs, speaker diarization, and speaker focus are included as standard features. This architecture is relevant to cybersecurity and compliance requirements in regulated industries.
The Arabic-English Bilingual Model in Clinical Settings
In health technology environments across MENA, English drug names, procedures, and dosages frequently appear within Arabic speech. Generic models mishandle these terms, and the resulting errors can enter patient records. The bilingual medical model accurately transcribes ICD-10-CM codes, drug names, dosages, and clinical shorthand regardless of which language carries them.
“This was critical to achieving meaningful outcomes for customers across the region who kept describing the same challenge. 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.”
Katy Wigdahl, CEO, Speechmatics
Sully.ai, a clinical documentation platform expanding across MENA, evaluated the model against complex clinical audio including code-switching and dialect-heavy consultations.
“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.”
Patrick Nguyen, Head of Engineering, MENA, Sully.ai
Both the standard and medical variants of the Arabic-English bilingual model are available now. On-premises and on-device deployment options make both models viable for regulated environments where clinical and enterprise artificial intelligence applications are being built across the region.
