Did you try their enhanced models? We're using it for relatively high-quality audio files and their accuracy is better than the whisper small.en model. More importantly, their word level timestamps is worlds better than whisper.