If you have a relevant text corpus (e.g. previous transcripts of the same person), you could use some Markov-type modeling/analysis to verify the transcription or find the most suitable alternative in the list.