Now you’re talking as Speechmatics wins California deal
Built on NVIDIA AI infrastructure, the collaboration combines best in class medical-grade speech models with autonomous agent workflows to deliver AI receptionists and clinical scribes that handle real operational tasks and deliver tangible ROI across clinical settings.
The partnership arrives as global healthcare faces acute staffing shortages and mounting administrative costs.
Ahmed Omar, Founder and CEO, Sully.ai said: “We needed best-in-class speech models that work in real clinical environments: complex medical terminology, fast overlapping dialogue, accents, imperfect audio, not just clean test clips.
“Speechmatics has been the most responsive provider with solutioning for us, and we’ve seen them handle medications better on our troublesome audio than any competitor.”
Speechmatics’ speech-to-text technology delivers industry-leading accuracy across 55+ languages, with specialised medical models trained on over 16 billion words of clinical data.
Sully.ai is transforming healthcare operations with autonomous AI agents that handle mission-critical workflows across multi-doctor practices and large provider networks.
The company’s full suite (from voice AI receptionists to clinical scribes) is built on an agentic operating system designed for the complexity of real healthcare environments.
Speechmatics and Sully.ai plan to expand into new global markets including the Middle East following the launch of an English-Arabic bilingual model in early 2026. Bilingual, code-switching conversations are expected to be a defining requirement for voice automation in care delivery and patient access within this region.
Speechmatics’ Arabic capabilities are designed to perform across Modern Standard Arabic as well as Egyptian, Gulf, and Levantine dialects, supporting consistent performance across varied speakers and accents.
Trained on over 16 billion words of medical conversations, clinical documentation, and healthcare interactions, the models deliver keyword error rates 5-20% lower than evaluated competitors on medical test sets across most languages. This training scale enables the models to distinguish between “hypertension” and “hypotension” in noisy emergency rooms, understand pharmaceutical names with regional accents, handle overlapping clinician-patient speech, and parse medical abbreviations, drug dosages, and ICD-10 codes, all while maintaining near-batch accuracy at sub-second latency.
Katy Wigdahl, CEO, Speechmatics said: “High-accuracy, low-latency speech recognition is a core enabler for autonomous agents that ‘actually listen’ and operate safely in mission-critical environments.
“Together with Sully.ai, we’re enabling healthcare organisations to deploy ambient scribes and AI agents across multiple languages and global markets without compromising on quality, security, or speed.”