Analysis of Transcriptions Using Octra – A Pilot Study

Authors: Christoph Draxler

Abstract:

Octra is a web-based editor for orthographic transcription of spoken language recordings. For this pilot study, 44 political speeches from Italy and Germany were partly pre-processed by automatic speech recognition and then corrected manually, and partly transcribed from scratch using Octra. We report the word error rate, and we propose time-based and timeless transcription factors to capture the effort to perform the orthographic transcription, and we present a visualization to gain insight into how transcribers actually perform the task.