Really looking forward to seeing people's solutions. I've not found a way to get around the slightly strange number / abbreviation formatting with CTC & post processing. I'm guessing the top performers have gone for encoder-decoder models like Whisper.
All the best :)
Really looking forward to seeing people's solutions. I've not found a way to get around the slightly strange number / abbreviation formatting with CTC & post processing. I'm guessing the top performers have gone for encoder-decoder models like Whisper.