Speech Recognition

High quality, real-time speech recognition tools now provide a viable and economic alternative to conventional keyboard-based methods for live subtitling. SysMedia has been at the forefront of the introduction of this technology for television subtitling.

The use of speech recognition for subtitling does not mean recognising the voice of the TV presenter - the technology is not yet good enough to deliver sufficient accuracy. Instead, a trained "re-speaker" listens to the programme, and re-dictates a slightly edited transcript of what is said into a speech recognition system that has been trained for their voice. The re-speaker is trained to achieve high accuracy and consistency.

WinCAPS integrates speech recognition for both live and offline transcription through a special feature called SpeakTITLE. This feature does more than just interface to all the commonly used speech recognisers, it also adds quality enhancement to the recognised text by normalising text presentation and performing linguistic post-processing. When subtitles are presented in a scrolling word-by-word format, SpeakTITLE smooths the rate of text delivery so it is easy to read.

As well as its use for live subtitling, speech recognition is used for offline preparation and SysMedia's SpeechTime option for WinCAPS allows timecodes to be captured during a transcript of a recorded programme in such a way that a viable first-pass subtitle file is obtained.

Our speech experts are available to provide training on how to achieve best performance from speech recognition technology and thereby to reduce subtitling costs and increase productivity.

<< Newsroom Interface

Ingest and Capture >>