TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Text to Speech from voices with ~15 minutes of Audio from YouTube videos

5 点作者 sentdex将近 6 年前

2 条评论

ksaj将近 6 年前
Elon sounded like he was talking backward, but I think you're definitely on the right path. Really interesting.
sentdex将近 6 年前
Using transfer learning on top of a DCTTS model (deep convolutional text to speech), I wanted to see how quickly one could recreate voices remotely convincingly.<p>TLDR&#x2F;W, using ~15 minutes of audio and about 1.5 hours of training, I was able to create what I think are pretty good examples of voices of myself, Donald Trump, Obama, Musk, and Joe Rogan.<p>None are perfect, and very much still a work in progress, but maybe something you might want to note that exists now (and has for years).<p>Even if you don&#x27;t post videos of yourself on YouTube, your audio is almost certainly stored, tagged by your name, by Google (Assistant), Apple (Siri), Amazon (Alexa), and probably many other providers.