Cause: Something to do with how Blender finds primary display in Windows environment and that data is saved in the startup blend file???

Fix: A bug report (https://developer.blender.org/T34962) has been opened for a very long time, but we need to further narrow down the root cause in order to fix it.

Workaround: When Blender opens in the wrong display (usually after an upgrade or after saving a new startup.blend file), you have to change your Windows Display Settings, then resave your default startup file, then reset your WIndows Display Settings back to your normal configuration. When you reopen Blender, it should open in the primary display device.

NOTICE: Depending on your display dimensions and configuration, fixing Blender may cause you to have to reset some custom settings (like reorganizing your desktop shortcut icons). The inconvenience of having to reset the locations of a few icons after the process usually far outweighs the inconvenience of the blender bug.

Instructions:

1. Go to your desktop, right-click on your desktop and choose Display Settings
2. Go to Multiple Display settings and change from Extend to Duplicate Display, apply and click Keep Settings. Minimize settings window.
3. In Blender, resave your default.blend file by going to File -> Defaults -> Save Startup File and then close Blender.
4. Go back to Display Settings again (maximize the window or if you closed it: go to your desktop, right-click on your desktop and choose Display Settings).
5. Go to Multiple Display settings and change from Duplicate to Extend Display, apply and click Keep Settings. Close settings window.
6. Open up Blender and it should now open in the primary display device.

1. Record an audio track of character dialogue.
2. Analyse the audio track using speech-to-text artificial intelligence.
3. Receive speech-to-text results, but with the time offset information for words and phonemes.
4. Import those encoded results into a Blender 3D animation timeline.
5. Blender uses those results to match phonemes with timeline.
6. Mouth shape from character pose library is selected based on phoneme and timeline.

While this sounds like a dream (because it would be), I actually think the pieces for this are already out there. With Google Speech API, I can post my audio file to the AI and get reliable speech-to-text conversion with word confidence scores. If in our Python script, we set:

enable_word_time_offsets=True

we get the text results with time offsets for each word. I'm going to check with Google, but I bet there is a debug flag available to get the offsets for time offsets for each individual phoneme. Why can't we use that data to re-associate the words with our timeline in Blender?

Working from the other end of the pipeline, I see a Papagayo product that puts text into mouth shapes, and I see a Blender addon called Automatic Lipsync that puts the Papagayo data into Blender.

Mission: Don't we now have the technology to put ALL of these together into either a addon plugin, or better yet core?

I'm fairly new to Blender, so this is going to be above my skill level - but folks - although ambitious I see no reason why this isn't possible? Where do I begin to make this happen or what would be the most appropriate forum to further the discussion?

Resources:

A primer on using Google Cloud Speech API and how speech recognition works.

A page with instructions and example Python code for processing audio with time offsets.

The official page for Papagayo.

The official page for Lip Sync Add-on.

And of course, Blender 3D!