Does the length of the initial-prompt affect the transcription time of the first window?

[Jemtaly]

Hello everyone, I am trying to make a real-time transcription application based on whisper. The basic implementation idea is to cyclically transcribe the last five seconds of speech (in the specific implementation, some cutting and adjustments will be made based on the intervals between words, which will not be detailed here).

In the strategy I am currently using, in order to ensure the coherence of the transcribed content, I will use all the previously transcribed content as the prompt for the current transcription window. However, I found that as the running time increases, although the length of each audio piece is basically They are all around five seconds, but the transcription speed will get slower and slower. This problem will not occur if I always keep only the most recently transcribed n words as prompts (but this sometimes affects the accuracy of the transcription).

So, I want to know, will the length of prompt affect the transcription speed? How is it affected?