G2

Verified User in Broadcast Media

Enterprise (> 1000 emp.)

10/26/2026

Business partner of the seller or seller's competitor, not included in G2 scores.

"Speechmatics Real-Time Engine: Rock-Solid Transcription for Live TV Captioning"

5/5

What do you like best about Speechmatics?

We run Speechmatics’ real-time engine as the transcription core of a live captioning service for broadcast TV. Each channel gets its own Speechmatics container, deployed both in the cloud and on our own Kubernetes hardware. After months in production and a lot of testing, it has earned its place in our stack. Review collected by and hosted on G2.com.

What do you dislike about Speechmatics?

The least helpful thing about Speechmatics is how much its high-accuracy model depends on the hardware underneath it. The official sizing guidance is demanding, and on an older CPU generation we watched it fall behind real time, which is fatal for live captions. Recent CPUs ran it comfortably, but you only find out which side of that line you’re on by benchmarking yourself. Self-hosting also brings license management and usage reporting that a hosted API would handle for you. And the engine holds the end of a sentence until speech resumes, so when a live feed goes quiet, the last few words can sit on hold until audio comes back. Review collected by and hosted on G2.com.

What problems is Speechmatics solving and how is that benefiting you?

Speechmatics solves the hardest part of live captioning: turning unscripted broadcast speech into accurate, readable text within a few seconds, at a scale where human captioners are impractical. Broadcast audio is noisy, fast and full of accents and proper nouns. Speechmatics handles that well and returns punctuated, formatted text that is ready to put on screen. It also solves where the transcription runs. Because the engine is a container we host ourselves, audio never has to leave our own or our customers’ networks. That matters for broadcasters with strict security requirements and for isolated sites with no internet access.

For us, this means we can offer automated live captions as a product, not a staffing problem. We launch captioning on a new channel in minutes, run it around the clock, and deliver captions to air about four to five seconds behind the speech, consistently, for days at a time. Its light resource use lets us run many channels on modest hardware, which keeps the cost per channel low. Built-in health checks let our orchestration recover from problems on its own, without anyone being paged. And since one engine works in the cloud and on-premises, we built the captioning service once and deploy it in both places. Review collected by and hosted on G2.com.