i mumble, so dictation misses me. now every voice note i send my ai is saved with its transcript, building a corpus to fine-tune a speech model on my voice
Recipe No. 4
Turn Everyday Voice Notes into a Training Set for Your Own Speech Model
Serves People with an accent, a mumble, or a speech pattern that dictation keeps getting wrong
Save every voice note together with its transcript into one private folder, so the pairs become a training corpus for a speech-to-text model fine-tuned on his voice.
You will need
Method
Ask the agent to save each voice note's audio and its transcript side by side, with a date.
Fix wrong transcripts when you notice them; each correction makes a better training pair.
Have it write a small local training script you can point at the folder once you have a few hours of audio.
Fine-tune, then compare old versus new accuracy on a few notes it never saw.