The first step in transcript alignment is to generate a speech-to-text transcript. IDOL Speech Server then compares the original transcript with the speech-to-text transcript. For best results, run the speech-to-text task using a transcript language model with a suitable interpolation weight. The suggested range of weighting is between 0.5
and 0.9
. Use the higher value if the transcripts are almost exact.
For more information about how to perform speech-to-text, see Speech-to-Text.
|