In the diploma thesis, we analyzed intelligent tools in Adobe Premiere Pro, specifically speech transcription, scene edit detection, and generative video extension, evaluating their effectiveness, advantages, and limitations. Testing was conducted on diverse video clips, including dialogues in clean conditions, clips with noise, dynamic and static scenes, using metrics such as Word Error Rate (WER) and Diarization Error Rate (DER) for transcription, Precision (precision), Recall (recall), and F1 Score for scene transition detection, and visual scoring (on a scale of 1–5) for generative extension.
The results show that transcription achieves high accuracy (0% word error rate) in normal conditions but performs poorly with noise (up to 277.3% WER) and accents (24.5%). Scene edit detection is accurate (precision up to 99.5%) but less reliable for soft transitions. Generative extend is effective for simple clips (rating 4–5) but causes distortions in complex scenes (rating 2–3.5).
These functions save 50–80% of time compared to manual editing but are limited by reliance on cloud processing and inadequate support for multiple languages. We conclude that the tools significantly enhance productivity but require further improvements in noise handling and complex content processing, positioning Premiere Pro as a leader in AI-supported video production.
|
