Granada, November 14-16, 2022

Albayzin 2022 Text and Speech Alignment Evaluation, task 1: Subtitles alignment

Respeaking, as carried out by a respeaker, is the process of repeating what is heard into voice recognition software, which is trained to that specific individual’s voice and pronunciation. The IberSPEECH-RTVE 2022 Text and Speech Alignment Challenge aims to evaluate the text to speech alignment systems on the actual problem of synchronising re-speaking subtitles with the spoken audio. The task will assess the state of the art of offline alignment technology. The purpose is to provide the subtitles without delay for a new broadcast. In this task, participants will be supplied with the subtitles as they originally appeared on TV, including the start and end timestamps of each subtitle. Participants must provide an output with the exact same sequence of subtitles but with the new start and end timestamps for each subtitle. It should be noted that re-speaking subtitles often differs from the actual spoken words. If the speech is too fast the re-speaker tends to suppress words (deletions) or even paraphrasing, which introduce a new level of difficulty in the alignment process. The performance will be measure by computing the time differences between the aligned start and end timestamps given by the alignment systems and the reference timestamps derived from a careful manual alignment.

The calendar for the Albayzín Text and Speech Synchronization evaluation is:

  • May 3rd, 2022: Registration opens
  • May 3rd, 2022: Release of the training and development data
  • September 4th, 2022: Registration deadline.
  • September 5th, 2022 Release of the evaluation data
  • October 16th, 2022: Submission deadline
  • October 24th, 2022: Results distribution to the participants
  • October 30th, 2022: Paper submission deadline
  • November 15th, 2022:Iberspeech 2022 Albayzin Evaluations special session in Granada
More details in the evaluation plan