Whisper API stutter and erring like LLMs

I’m using Whisper to transcribe some non-English audios and it showed this super weird stuttering in its output, like repeating a word for many many many many times, which is actually a typical bug for unmature language models. I guess they use some kind of LLM to boost their performance. I would really hope OpenAI can offer some more precise transcription services because we can feed the raw transcript into GPT ourselves

1 Like