Recommend passing the speech-to-text narration through a round of GPT4 API to correct for any transcription errors (use some prompt giving context that it's speech to text)