Voice AI vendor Speechmatics cites independent Pipecat benchmark for lowest word-error rate, discloses no insurance terms of its own

Voice AI vendor Speechmatics says an independent benchmark by Pipecat found its speech-to-text service had the lowest error rate among 12 services tested. "In independent testing by Pipecat (as of August 2026), Speechmatics returned a 1.07% pooled word error rate, the lowest of the 12 services benchmarked, including Deepgram, AWS, and Azure," the company's website states.

The claim appears alongside the company's new Melia speech-to-text model, which Speechmatics says is "Available today in production preview, starting with batch transcription." The vendor also markets pairing its transcription API with "LiveKit's flexible agent framework to build world-class agents."

Speechmatics reports "56+ languages" of coverage and says it does not retain customer audio by default: "We don't log your data as standard," with deployment available "on device, on prem and in the cloud." For regulated sectors, the company markets a medical model it says cuts "errors on key terms by up to 50%" for "ambient scribe and dictation."

The company states it holds multiple compliance certifications: "SOC 2 Type II-certified," "ISO/IEC 27001:2022 accreditation," and it says it is "Fully compliant with the Health Insurance Portability and Accountability Act (HIPAA)."

On AI Dispatch's five-dimension trust-gap rubric, Speechmatics is one of a small number of entities in AI Dispatch's index whose central performance claim rests on an independently sourced benchmark -- the Pipecat test -- rather than a self-reported figure. The company discloses no runtime-governance mechanism, insurance coverage, indemnification terms, liability cap or outcome-linked payment structure for its own service, according to AI Dispatch's review of its published material.