8th International Conference on Spoken Language Processing

Jeju Island, Korea
October 4-8, 2004

An Efficient Repair Procedure for Quick Transcriptions

Anand Venkataraman, Andreas Stolcke, Wen Wang, Dimitra Vergyri, Jing Zheng, Venkata Ramana Rao Gadde

SRI International, USA

We describe an efficient procedure for automatic repair of quickly transcribed speech. Quickly transcribed (QT) speech, typically closed captioned data from television brodcasts, usually has a significant number of deletions and misspellings, and has characteristic absence of disfluencies. Errors of this kind often throw an acoustic model training program out of alignment and make it hard for it to resynchronize. At best the erroneous utterance is discarded and does not benefit the training procedure. At worst, it could misalign and end up sabotaging the training data. The procedure we propose in this paper aims to {em cleanse/} such quick transcriptions so that they align better with the acoustic evidence and thus provide for better acoustic models. Results from comparing our transcripts with those from careful transcriptions on the same corpus, and from comparable state-of-the-art methods are also presented and discussed.

Full Paper

Bibliographic reference.  Venkataraman, Anand / Stolcke, Andreas / Wang, Wen / Vergyri, Dimitra / Zheng, Jing / Gadde, Venkata Ramana Rao (2004): "An efficient repair procedure for quick transcriptions", In INTERSPEECH-2004, 1961-1964.