• 1 Post
  • 27 Comments
Joined 3 years ago
cake
Cake day: July 5th, 2023

help-circle
  • Yes, but…

    The reason why iOS asks you to read specific things out loud is because of a thing called alignment - basically, it needs to map exactly sounds to exact phonemes. So if it’s making you say “apple”, it’s recording how you say “ae”, “puh”, and “l”.

    When you expect it to take any random file, it needs to know specific things like, what’s the text, what is the time frame, when did the speaker speak in a normal voice vs a higher pitch or speed voice. Taking all of this information and created a voice is called “forced alignment” and that’s a field that’s well studied but not implemented on mobile phones as yet.

    An alternative for you, if you want, say, Snoop Dog to be your Siri voice, is to actually use an AI generated Snoop Dog voice to say all the things that the Personal Voice feature expects and just play them out loud from your computer when setting up Personal voice. This assumes you use one of the many paid Snoop Dog AI voice companies out there.

    Another way to get the voice you want is to just take a recording of someone, write the text of what they said and then feed it to a tool like the Gentle Forced Aligner - https://github.com/strob/gentle

    This will give you enough material to then take it to AWS or Microsoft’s AI voice creator tools to create an AI voice. Then you can use that AI voice to say the same things that iOS Personal Voice wants you to say and again, play the correct recording when activating Personal Voice.

    All of this assumes that is your end goal - Snoop Dogg as your Siri. Anything else, please let me know and I’ll help you with what you’re trying to achieve.


















  • It’s a corp. It absolutely can and will and has many times in the past made claims and not followed on them.

    One example - they said Google Photos will remain free forever for high quality photo uploads. But the moment they reached critical mass of photos needed to train their AI models, they switched to a paid model for higher quality photos. Leaked internal discussions said as such. They chose to make claims and then backtracked on them.