About this Video

• Video Title: Whisper OpenAI vs Whisper CPP vs Whisper Const-Me in Subtitle Edit 3.6.12
• Channel: David Mbugua
• Speakers: David
• Duration: 00:10:38

Introduction

This video compares the speed and accuracy of three different Whisper AI models—OpenAI, CPP, and Const-Me—for generating subtitles within Subtitle Edit 3.6.12 software. The primary focus is on speed, with a secondary consideration of accuracy.

Key Takeaways

• Whisper Const-Me (GPU): Fastest option, especially on Windows systems with a compatible graphics card. A two-minute clip took approximately four seconds to process. Subtitle accuracy is good, requiring minimal edits.
• Whisper CPP: A middle-ground option in terms of speed. Processed a two-minute clip in around 15 seconds. Subtitle quality is comparable to Const-Me.
• Whisper OpenAI (Python): Significantly slower than the other two options; processing time is roughly double the video length. While some users report GPU usage, this wasn't the case in this demonstration. Accuracy is similar to the faster options.
• Speed Variation: Processing times varied significantly between the models, highlighting the importance of selecting the right model based on the user's needs and system capabilities.
• Subtitle Edit Functionality: The video demonstrates how to access and utilize the audio-to-text features within Subtitle Edit 3.6.12, showcasing the ease of selecting different Whisper models.

The video concludes that for speed, Whisper Const-Me is the best option on Windows systems with a compatible graphics card, followed by Whisper CPP, with Whisper OpenAI being the slowest. While all three offer similar accuracy, requiring minor edits, the significant speed differences make choosing the right model crucial depending on project needs. The presenter also notes the potential for improved accuracy and timestamp precision in future updates.