Forum user Velociraptor Alfa pointed me today to THIS amazing AI using Google's "Collab" system. "John Hammond Synthesis" apparently used the untouched John Hammond recordings (that were available thanks to the "alpha" extraction we did years ago that I used for a couple variations of the Jurassic Time Hammond Memoir) that have been available on my site for some time now. It was created by forum user James_UK, AI Clone Voices on YouTube.
https://colab.research.google.com/drive ... ueQmxlnM04
How To Use:
At the link above, as you scroll down make sure you "run" each cell (at the top left of each one, when you hover over the brackets [ ] and it turns into a "play" style button) to download what is needed to use the AI. Or, to put it more step by step, "run" each cell under each heading below, the first line of each heading where the "play" should be right beside it:
"Install Tacotron2 and Waveglow"
!nvidia-smi -L
"Download Models"
force_download_TT2 = True
"Initialize Tacotron2 and Waveglow"
%matplotlib inline
Then under that: # Load Tacotron2
Then under that: # Load WaveGlow
(Note: You will have to go through the process above every time you close the tab/browser, otherwise you'll only have to do what comes next below.)
Once you reach the "Start playing around with the model" cell, change the text within the quotes where it says "Change the text here to make John Hammond say what you want" to make it be whatever you want. Run that same cell after you have changed the text to what you want, and it will produce two sound files below it; a standard one and a "denoised" one of your choosing. You can play each one and also download each one by clicking on the 3 dots on each file.
Some tips:
-Apparently this works better with shorter sentences at a time, so if you really want him to say something specific well it may be best to split it up then edit it together.
-Sometimes if you simply run the cell again you may get a different/better result. Download each one to see which "take" is best.
-Certain words can sometimes be hard for him to say correctly or how you want, so you may want to try a phonetic approach. For example: if you use the word "juniper" he says it like "june-EYE-per". So to have him say it correctly, I changed the actual word to "jew-nip-er".
-Periods, commas, and hyphens can help in pacing of what is said; even if they aren't technically correct.
I have to say, I think this AI is astounding and shocking all at once. If someone could create an offline version of this, that would be awesome! For now, enjoy it as I have! Hours of fun to be had with this!
And please... try not to make him swear TOO much.
I think once the AI gets better at recreating Attenborough's voice you could attempt to intermingle the actual performance with the reconstructed lines. I guess a third option that does not require the use of AI is finding a good impersonator. There are a surprising amount of talented voice actors who can imitate Richard's brother David Attenborough's voice. But I've come to realize that Richard's performance is so much harder to do because of his cadence.
