Speechmatics

I had a dream... from the moment we got a voice agent running on our tech in 2024, I knew the perfect demo I wanted to deliver. This week, we made it happen... Picture the scene: a stage, the biggest screen you’ve ever seen behind me, 500 people in the audience; decision makers across finance, insurance, airlines, accounting and countless other industries that make up boost.ai’s customer base. We’d had the keynotes from Jerry and Ankur. David had just introed my panel in his multicoloured shoes and off we went. What ensued was a natural, mildly chaotic and genuinely funny panel between 3 voice agents (and myself) talking through the latest innovations in voice tech that are driving the push from agentic chat to agentic voice at a blistering pace. I’ve been lobbying the demo team at Speechmatics to let me do this for so long, but it’s always been a no because of the risk and challenge of doing it live and above all dealing with the interruption risk from the mic and speaker feedback threatening a doom loop where agents cut themselves off. This time was different. This time we used a combination of the latest tech to make sure things worked. Speaker ID meant the agents would recognise their own voices so couldn’t interrupt themselves Diarization meant each knew when the other was speaking A huge lexicon and custom dictionary meant we could talk about little known products, made up names and random towns Short utterance training meant that one word answers were fully understood And alphanumeric work meant that emails and number plates could be understood flawlessly. Getting this to work on stage is so much harder than delivering a smooth agentic call, so the fact we could should be a lesson to all that now is the time to bring voice AI into your workflows. Next time it will be multilingual too! A huge thank you to the Speechmatics team for working so hard on the demo and to the Boost team ( Samantha, Axel) for trusting that we could deliver where OpenAI had so spectacularly failed only a few weeks ago at their OpenAI DevDay. I can’t wait for the next one.

Keep confidential conversations, confidential. Speechmatics On-Device processes speech locally, so notes from patient-doctor consultations can stay on the user’s machine. For teams building clinical documentation tools, that means sensitive data isn’t being sent elsewhere for processing. Keep patient conversations confidential. Request access to On-Device.

Running transcription offline shouldn’t mean compromising on quality ❌ Speechmatics On-Device processes speech locally, with accuracy within 5% of our cloud standard models. For medical scribes, sensitive audio can stay on the device while you maintain the high-quality transcription that healthcare demands. Keep sensitive speech data local, without risking patient safety from inaccurate transcripts.

Speech-to-text that runs entirely on the device, no cloud, no network required. 🎙️ Speechmatics’ On-Device brings production-grade transcription local: processed on the user's machine, with near-cloud accuracy across 55+ languages. It’s already used by customers in production today.

Patient care doesn’t stop when there’s no internet connectivity. Neither should your medical scribe. Speechmatics On-Device runs locally, with no internet required for transcription. That gives clinical documentation products a way to keep working in environments where connectivity is limited, unreliable, or unavailable. Keep transcription running without relying on the cloud. Request access to On-Device.

The appointment is confidential. Is your transcription? Speechmatics’ On-Device lets you use speech-to-text while keeping audio and transcripts local to the user’s machine. That means fewer questions around cross-border audio transfer and cloud data handling, without compromising much on transcription quality in 55+ languages. Give privacy-conscious customers another way to say yes. Request access to On-Device.