That would mean 2 things, maybe the customers accent and poor audio input. This kind of transcription error is common with speech-to-text models, especially with accents, unclear audio, or similar-sounding words. In your assistant config, you can define expected phrases in the intent parser or custom logic to correct or adapt to common errors. You could write: - If transcript includes "mary" → assume it's possibly "married" - If "daddy's" appears → assume it might mean "thirties" - Using noise suppression background noise - Encouraging clearer responses (e.g., "Can you say that again clearly?")