The demand for TV programs subtitling has grown fast in the past few years throughout Europe. The path of manual subtitling is no longer feasible, due to the cost of the process, both in terms of time and personnel. As a result, broadcasters and subtitling companies are seeking for subtitling alternatives more productive than the traditional manual process.
Respeaking – is a technique in which a professional listens to the source audio and re-speaks it to a speech recognition engine which transcribes it – is consolidating as the main subtitling technique employed for live and pre-recorded broadcast productions.
Assisted Subtitling is an emerging technique, consisting on the application of Speech Recognition to automatically generate a transcript of a programme’s soundtrack without the need of a respeaker, and to use this as the basis of subtitles.
Large Vocabulary Continuous Speech Recognition (LVCSR) is the technology underlying Respeaking and Assisted Subtitling, and it is proving to be useful for such a purpose. Nevertheless, most respeaking and assisted subtitling applications rely on commercial large vocabulary continuous speech recognition (LVCSR) software. Such commercial systems have two main limitations:
- they have been developed to target the main language markets (i.e. English, Spanish, French, German, Mandarin Chinese or Japanese among some others) and thus, are not available for many EU languages
- they have been designed for dictation applications and, as a consequence, their performance is too poor to be used in certain tasks, like Live Subtitling, and also their language coverage may be limited in specific domains.
SAVAS aims to acquire, share and reuse audiovisual resources of broadcasters and subtitling companies to develop a brand new Automatic Speech Recognition (ASR) technology for Multilingual Live Subtitling, specifically tuned to the needs of the Broadcasting and New Media Industries. An initial set of 7 languages will be developed: Basque, Spanish, Italian, French, German, Portuguese and English. A wide set of specific Dictionaries for LVCSR will be created, solving current limitations for the following domains:
a. Broadcast News
c. Interviews and Debates
Thanks to the SAVAS Data Collection and Acquisition methodology, based on the combination of automatic and collaborative approaches, the creation of new Languages or new Dictionaries will be automatized as much as possible, thus reducing the creation effort and time-to-market.