On automatic cross-lingual subtitle timing

In the last years automatic speech recognition (ASR) technologies are applied as assistant technologies for hearing-impaired people. ASR technologies typically produce some kind of subtitles by two ways: i) directly by the automatic speech recognition [1], [2] or ii) provide the timing for already existing transcription [3], [4], [5]. Both mentioned solutions are constrained by the fact they produce the subtitles in the same language as is the language of audio recording. In this paper we address the task of cross-lingual subtitle timing (audio data language differs from the language of given subtitles). We define few scenarios, which differ in the available input data. For each scenario we propose a solution, evaluate preliminary results and propose how to solve main weaknesses.