Automatic subtitles

Automatic subtitles are caption files produced without manual typing, mapping each text segment to a time interval in the audio. The two standard formats are SRT (the older one, accepted everywhere) and VTT (its web-native equivalent, read directly by browsers through the HTML5 track element).

Two ways to generate them

There are two routes to automatic subtitles. The first starts from a recording and transcribes it with speech recognition: that is YouTube's approach, with the transcription errors everyone has seen on proper nouns and technical terms.

The second starts from the text. When the audio is generated by text-to-speech from an article, the exact text is known in advance and every word carries a timestamp. Subtitles then become a simple formatting exercise over data that is already exact: splitting the text into lines of reasonable length and aligning them with the speech timing. No transcription errors are possible, because nothing is transcribed.

What publishers use them for

Three main uses. Social video first: an audio article repackaged as a video for social networks needs subtitles, since a large share of social video is watched with the sound off. Accessibility second: captions make audio content usable by deaf and hard-of-hearing audiences, and they are a standard requirement in accessibility guidelines for any time-based media. Situational flexibility last: in loud or silent environments (public transport, open-plan offices), the caption stands in for the sound.

For a newsroom, the case for automation is the same as for audio itself: covering the entire output, not just flagship pieces. The WeDispatch subtitles feature generates SRT and VTT files for every audio article, ready to attach to a video or a player.