Advanced search


Emilio Granell. Combining Outputs from Multiple Handwritten and Speech Recognition Systems for Transcribing Historical Handwritten Documents. 2015. PRHLT seminars

Given that combining the outputs of different recognition systems will generally improve the recognition accuracy, we have been working on the multimodal combination of automatic speech (ASR) and handwritten regonition (HTR) systems, in order to obtain a better draft transcription of historical documents. We will present a new combination technique based on confusion networks. In a first experiment, results on two different sets of data (Cristo Salvador and Rodrigo), with different difficulty level, show that the proposed technique provides similar or better draft transcriptions than a previously proposed approach, in a multimodal and iterative scenario. In a second experiment, results on unimodal combination with different optical (for HTR) and acoustic (for ASR) models, and on multimodal combination, show important error reduction on both, the WER and CER levels over the HTR baseline. PRHLT