Improving Lightly Supervised Training for Broadcast Transcription
MetadataShow full item record
Long, Y., Gales, M., Lanchantin, P., Liu, X., Seigel, M., & Woodland, P. (2013). Improving Lightly Supervised Training for Broadcast Transcription. Interspeech 2013, 2187-2191. http://www.isca-speech.org/archive/interspeech_2013/i13_2187.html
This paper investigates improving lightly supervised acoustic model training for an archive of broadcast data. Standard lightly supervised training uses automatically derived decoding hypotheses using a biased language model. However, as the actual speech can deviate significantly from the original programme scripts that are supplied, the quality of standard lightly supervised hypotheses can be poor. To address this issue, word and segment level combination approaches are used between the lightly supervised transcripts and the original programme scripts which yield improved transcriptions. Experimental results show that systems trained using these improved transcriptions consistently outperform those trained using only the original lightly supervised decoding hypotheses. This is shown to be the case for both the maximum likelihood and minimum phone error trained systems.
lightly supervised training, speech recognition, confidence scores
The research leading to these results was supported by EPSRC Programme Grant EP/I031022/1 (Natural Speech Technology).
This record's URL: https://www.repository.cam.ac.uk/handle/1810/245711