A transcript can be produced in several ways - either you listen to the show and write down what you’ve heard (which will take hours to do) and then clean it up, or use a tool or open-source speech-to-text model like Whisper to generate a transcript automatically. The audio processing step is split into three steps - transcribing the audio (converting speech to text), aligning the transcription (determine the exact timing for each segment of the transcript), and diarization (identifying and assigning speaker labels). _cleanup_gpu_memory ( model ) return result def _align_transcription ( self , audio , result : Dict ) -> Dict : """Align transcription with audio."""