University of Twente Student Theses

Login

Monitoring van een ASR-systeem : het gebruik van interne confidencewaarden en teletekstondertiteling om de Word-Error-Rate van een herkenning te voorspellen

Brouwer, L.J. (2006) Monitoring van een ASR-systeem : het gebruik van interne confidencewaarden en teletekstondertiteling om de Word-Error-Rate van een herkenning te voorspellen.

[img]
Preview
PDF
978kB
Abstract:Monitoring in automatic speech recognition (ASR) implies that the Word-Error-Rate of the recognition can be predicted without human intervention. Human intervention in this case means that a human will listen to the speech and write down what is said. This is called transcribing. Measuring is then done by comparing the recogniser-output with the transcription. The Word-Error-Rate(WER) is the measure of how well a recognition was done. The proposed monitoring will use internal likelihoods and closed caption to calculate new measures that estimate the WER. This monitoring is developed for the Spoken-Document-Retrieval system that is developed at the University of Twente. Here a web-portal is online where the daily 20.00 `o clock broadcast-news is automatically recognised. The user can then search in the textual representation of the audio, and view the video of the corresponding fragment. For this system, monitoring should indicate the quality of parts, or all of the recognition. If monitoring shows that some parts are badly recognised, actions can be taken manually to fix the problem.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Human Media Interaction MSc (60030)
Link to this item:http://purl.utwente.nl/essays/57212
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page