Live speech transcription and instant translation have gone from a technological promise to an everyday tool. From conference halls and TV studios to courtrooms, software helps convert spoken words into readable text within seconds. Developers combine the convenience of a web application with the option of secure deployment directly on the client’s premises.
From early attempts to live transcription
The first speech recognition solutions in this team appeared around 2000, and in 2005 a system for Czech and Slovak called Newton Dictate from Newton Technologies was created. The impetus was the need for massive transcriptions for media monitoring (in Slovakia it was overseen by Slovakia Online), so the project was developed in collaboration with the university in Liberec. The result was a foundation to build on.
Today’s version handles real-time transcription of speech and can detect sentence boundaries, speaker changes, and direct speech. For English they report an accuracy of around 97%; for Korean, during a demo they jokingly spoke of "110%", underscoring the high quality. The system can not only transcribe but also translate between languages.
Conferences, television, and subtitles within four seconds
The technology has proven itself at dozens of conferences and today also powers live subtitling of TV broadcasts 24/7. In practice that means that within four seconds there must be not only an accurate transcript but also properly segmented subtitles that are readable even for the deaf. At one TV station (CNN Prima News) it turned out that live subtitling is faster and more accurate than the raw transcript itself.
Because it’s a web application, operators can tag speakers and the system learns to recognize their voices in subsequent broadcasts. Clarity is further improved by color-coding by person, so the viewer or editor quickly sees who is speaking. Such features reduce editors’ workload and speed up publishing outputs.
Security, deployments, and tools for newsrooms
In addition to the cloud, an on‑premise solution is available for institutions that need to keep data under full control. In the Czech Republic, the Ministry of Justice uses it to record and transcribe court hearings, as does the police; the spokesperson did not name one security service in Slovakia. The technology can also be deployed on an isolated notebook – the first client was Česká národná banka for classified meetings of the bank board – and in parliament it helps stenographers plan sessions thanks to integrated recording and a timer.
The system can transcribe, translate, subtitle, summarize, and search for keywords in streams, with everything remaining under human control. Operator corrections serve as valuable data for further improving the models, so the technology becomes more accurate year by year. Major clients include the Austrian news agency APA, which transcribes radio and TV broadcasts in German; the application also stores recordings and transcripts, enables full‑text search, and shares text with the newsroom in one click for quotes or quick summaries.
