When Leo cannot transcribe a word, does it just skip it? If so, is it possible to have it signaled into the transcription? Something like [illegible] would call attention to words that Leo skips, but the human user could guess by context, similarity with other texts, or skills.
Leo will give its best shot rather than skipping the word. We decided not to include an [illegible] sign as we didn’t want to kneecap the model. (The ultimate goal is to deliver not just passable but superhuman transcriptions!)
This is a good point.
My question is because I find, from time to time, that single words are missing from the transcription. I will provide samples.
Any examples would be great. And if you recognise any patterns for when it tends to happen that would be really helpful to know, too. Thanks!