The only thing that matters here is how good the transcription is. You absolutely have to save the recording. You also have to enable the user to connect to their own transcription service and preserve the recording for that if yours sucks or is not trusted. People have accents. Third party transcription vendors can sell data. Do not mess this up. Enable users to add their own trusted transcription.

If we want to give this to grandparents to save their stories, we can want to have the stories too. If we want it for ourselves, we have to trust it.