How to generate voiceover and subtitles for a (Powerpoint) presentation

Assuming you want good audio quality and want to avoid using YouTube for the generation of subtitles for some reason (time e.g.).

Guide for those in a hurry

1. Make a presentation. Avoid:
   1. Animations, make incremental slides for that
   2. If really necessary, you should extract the animated slides to a separate presentation later and render that out as a video.
2. Type the Audio track in the slide notes. You can check overall length by running notes-to-wav.ps1 and loading the audio into your favorite music player, like foobar2000
3. Export all slides as PNGs
4. Use some service to generate high-quality speech from the allnotes.txt, like Play.ht
5. Use your favorite video editor to arrange the slides according to the audio track, fine tune etc.
6. Render the video
7. You can use the vosk transcriber or, to make sure a local model is used
   1. pip install vosk
   2. have ffmpeg installed and in the path
   3. download the model from https://alphacephei.com/vosk/models
   4. use audio-to-srt-local.py -m <modelpath> <video.mp4>, this will write video-recognized.srt
   5. clean up/merge the subtitles using https://aegisub.org/downloads/
8. Import the srt into the video editor
9. Further adjust subtitles and timing
10. Re-Export subtitles
11. Profit.

Requirements

pip install python-dotenv vosk pyht

The scripts

notes-to-wav.ps1

This script generates several things for you:

• Slide[\n+].txt containing the notes text separated by slide
• Slide[\n+].wav containing the text synthesized by System.Speech.Synthesis.SpeechSynthesizer - this is helpful to keep track of the total talking time mostly
• allnotes.txt the whole text in a single file

audio-to-srt.py (deprecated)

This script generates a .srt from an audio track using the strategy described here

audio-to-srt-local.py

Generates a .srt directly from a video file using ffmpeg and vosk and a local model (download the model manually).