You need to line up the timestamps for each spoken sentence to each written sentence. Some data sources, like Amazon's "Whisper Sync" can do this already.