Published some notes on OpenAI Voice Engine TTS on my blog
It turns out even their flagship TTS custom voices were created using just 15 second samples from paid voice actors!
(On top of untold quantities of undocumented training data for the underlying model)
https://simonwillison.net/2024/Jun/8/how-voice-engine-works/
Simon Willison /
npub13v…4w5eu
2024-06-08 20:16:04
Author Public Key
npub13v97j0kknscwnf5pt87nsn7cxzxwfwl3dsu7ss8qsq7ukmqgwg8q84w5euPublished at
2024-06-08 20:16:04Event JSON
{
"id": "2324ecda6afd5f4dff3f6ee44d63ce1aa27abf3b66dce996dfbb27320b4151b9",
"pubkey": "8b0be93ed69c30e9a68159fd384fd8308ce4bbf16c39e840e0803dcb6c08720e",
"created_at": 1717870564,
"kind": 1,
"tags": [
[
"proxy",
"https://fedi.simonwillison.net/users/simon/statuses/112582365319399953",
"activitypub"
]
],
"content": "Published some notes on OpenAI Voice Engine TTS on my blog\n\nIt turns out even their flagship TTS custom voices were created using just 15 second samples from paid voice actors!\n\n(On top of untold quantities of undocumented training data for the underlying model)\n\nhttps://simonwillison.net/2024/Jun/8/how-voice-engine-works/",
"sig": "c1bd06befb79bf29279d2be7a0a7e73f523b3673caf57bb7949b60f3b7a1b37764b235c0e7d40a00c086ab672b34ae3ef7efb6fb8b76b2a503113248966f2b7a"
}