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。

Musika Fast Infinite Waveform Music Generation

1 点作者 hecanjog将近 3 年前

2 条评论

anigbrowl将近 3 年前
Better luck next time :-&#x2F;<p>This is like listening to music in the next room when you&#x27;re drunk or can&#x27;t sleep. Some of the features are identifiable (eg the 4 on the floor techno beat) but because it&#x27;s been trained on a large corpus it just ends up as a stew of spectral information.<p>The way to teach a computer genre music is to either train it on sections of good tunes, or iteratively train it on highly curated playlists, or (best imho) train it to generate sequences and then synthesize the results. As it is there&#x27;s no clear structure, no real tune, and the mix is a mess.
Rochus将近 3 年前
Interesting. But the quality seems not to be as good as e.g. <a href="https:&#x2F;&#x2F;www.deepmind.com&#x2F;blog&#x2F;wavenet-a-generative-model-for-raw-audio" rel="nofollow">https:&#x2F;&#x2F;www.deepmind.com&#x2F;blog&#x2F;wavenet-a-generative-model-for...</a>, neither in terms of sound quality nor in musical logic, structure and flow.