Lecture Notes: Turning Speech into Text with Whisper AI

Introduction

• Speaker: Kevin
• Topic: Transcribing speech to text using AI (specifically Whisper)
• Highlights:
  • AI transcribes better than most humans
  • Supports 96 languages
  • Effective with background noise and thick accents
  • Free and open source

Overview of Whisper

• Developed by OpenAI (same company behind ChatGPT and DALI2)
• Can be installed on a computer, but easier to use via Google Collaboratory

Setting Up Google Collaboratory

1. Accessing Collaboratory: Basic Steps

   • Use Google Drive and click on 'New'
   • Go to 'More' > 'Connect more apps'
   • Search and install Google Collaboratory

2. Starting Google Collaboratory

   • Click 'New' > 'More' > 'Google Collaboratory'
   • Name your file (e.g., 'Transcribe Audio')

3. Configuring Runtime

   • Go to 'Runtime' > 'Change Runtime Type'
   • Select 'GPU' for optimal performance

Installing Whisper AI

• Code Entry: Enter code in Google Collaboratory to install Whisper and FFmpeg from GitHub
• Run Installation: Click the run icon to install (takes about 23 seconds)

Uploading and Transcribing Audio/Video Files

1. Uploading Files:

   • Click the folder icon and drag in an audio or video file
   • Note: Files are temporary and will be deleted after runtime ends

2. Transcribing Audio:

   • Insert code to call the Whisper AI and specify the filename
   • Choose a model (tiny to large), medium model recommended for balance of speed and accuracy
   • Run the transcription command

3. Output Files:

   • Text file (TXT): Contains full transcript
   • Caption files (SRT, VTT): Include timestamps
   • Download these files if needed

Additional Features

• Whisper Parameters:
  • Use whisper-h command to view all parameters
  • Specify output location, translation options, language, etc.

Practical Use Cases

• Personal Application:
  • Used for YouTube video captions
  • More accurate than Google's auto-generated captions

Conclusion

• Whisper offers high-quality transcription with additional capabilities
• Encouragement to subscribe for more videos
• End of presentation