此刻的AI作曲，写出来的歌能够当短视频的BGM了
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/0637649595b4481a9e30d93a0a59e5ac~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=PLTUXtWKTxDw8ARdS04iEPFsils%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">毫无疑问， AI 的<span style="color: black;">显现</span>，让不少行业面临着技术革新，音乐圈子<span style="color: black;">亦</span>不例外。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">不仅人声模拟，在音乐创作这块儿， AI <span style="color: black;">亦</span>是卯足了劲，<span style="color: black;">各样</span>文本生成音乐模型是一个接着一个：</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">像是 OpenAI 的 MuseNet 、谷歌的 MusicLM 、 Meta 的 MusicGen ，还有前不久 Stability AI 家刚出来的 Stable Audio 等等等等。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">这还只是<span style="color: black;">有些</span>比较出圈的 AI 音乐模型，其他的不知名的<span style="color: black;">更加是</span>海了去了。</span></strong></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p26-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/33571acd3e2d46a785084e3d6de44de2~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=ZP6R22BWqAdzOY32%2FGPoGYJqCnc%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">这么多生成音乐的 AI 模型，它们主打的，都是一个让音乐门外汉<span style="color: black;">亦</span>能作曲，只要动动手会打字、会描述就 OK 了。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">这么一说，让没什么乐理知识的世超着实很心动，作曲咱不会，但文字描述可是咱<span style="color: black;">善于</span>的<span style="color: black;">行业</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">于是，<span style="color: black;">咱们</span>决定亲自试试<span style="color: black;">日前</span>市面上比较出圈的几款 AI 作曲模型，<span style="color: black;">瞧瞧</span>它们到底能<span style="color: black;">不可</span>实现从零作曲，以及写出来的曲子到底好<span style="color: black;">欠好</span>听、符不符合<span style="color: black;">需求</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">首要</span>出场的是 Stability AI 的新作曲 AI ：Stable Audio 。</span></strong></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/4d6946389dbe4720b882fff93516003c~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=NdnL0%2BC7DXAAPz1GNoqftiTCUJE%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">官方说是用了超过 80 万个音频文件去训练模型，里面像音乐、音效、单一乐器演奏等都有<span style="color: black;">包括</span>，<span style="color: black;">全部</span>数据集的时长加起来有 19500 多个小时。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">并且光靠语言描述， AI 就能生成最长 90 秒的音乐。</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">风格跨度<span style="color: black;">亦</span>是贼大，世超去它们官网听了下示例，有钢琴、架子鼓这种单纯器乐的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">还有<span style="color: black;">区别</span>流派<span style="color: black;">区别</span>风格的，<span style="color: black;">例如</span>民族打击乐、嘻哈、重金属之类的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">乃至</span>还能生成白噪音，像是一个餐馆里嘈杂的吵闹声， u1s1 听起来还蛮逼真的。</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">people-talk-in-a-busy-restaurant</span></span></strong><span style="color: black;">,差评,45秒</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">当然，官方<span style="color: black;">颁布</span>的肯定都是挑比较好的演示展示出来，到底用起来怎么样还是得亲自上手试试。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">于是<span style="color: black;">咱们</span><span style="color: black;">亦</span>注册了号，<span style="color: black;">瞧瞧</span>我这个音乐门外汉<span style="color: black;">经过</span>这个模型能创作出什么样的音乐来。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">因为</span>是刚发布，世超还花了好一会儿时间才进到 Stable Audio 的<span style="color: black;">运用</span>网页。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/4224db8de3574d66940d9a7edfd74b39~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=VqN0oaA6cgmOjVGIKrsLu8zVeVU%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">进去之后，<span style="color: black;">咱们</span>先让它生成一段 30 秒的贝斯 solo ， 112 个节拍，要 funk ，有律动一点。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/91bfc5386fd04e1eb9629e239599cd08~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=WqpVXeAYsee7U1QM92SqIo78boU%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">生成过程大概用了一两分钟，世超听了下结果，倒是有点出乎意料，是在弹贝斯没错，音乐风格<span style="color: black;">亦</span>挺准确，</span><strong style="color: blue;"><span style="color: black;">但<span style="color: black;">独一</span>的瑕疵<span style="color: black;">便是</span>这贝斯的音色不太清晰</span></strong><span style="color: black;">，像<span style="color: black;">指的是</span>弹和 slap 的中间态。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">接下来上点难度，乐器<span style="color: black;">繁杂</span>点，让它生成一段朗朗上口的流行舞曲，中间带着热带打击乐，要有欢快的节奏，适合在沙滩上听。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">这次 Stable Audio 有点小失误，虽然节奏挺欢快的，<span style="color: black;">亦</span>挺适合在沙滩蹦跶的，但提示词里的热带打击乐，我愣是没在这 30s 听出来。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">再让它生成一段摇滚曲风的音乐，<span style="color: black;">亦</span>是不出几分钟就搞定了，虽然听起来依旧不怎么清晰，但摇滚曲风以及电吉他、架子鼓的声音还是能听出来的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">整体体验下来，在音乐生成这块， Stable Audio 的表现确实<span style="color: black;">无</span>什么大错，偶尔还会有<span style="color: black;">有些</span>出乎意料的表现。</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">起码<span style="color: black;">针对</span><span style="color: black;">有些</span>想给短视频插背景音乐的创作者<span style="color: black;">来讲</span>，这个完全够用了。</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">并且这次， Stable Audio 还专门在时长上下了一点功夫，普通版<span style="color: black;">能够</span>生成 45 秒以内的音频，想要更长的话，就升级个 PRO 版，<span style="color: black;">能够</span>连续生成 90 秒。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/3ab59bfc1bff424d9b55d54ef6389c28~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=PZ7YAlsaLn%2B5v8o68grYTWBae7U%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">接下来上第二位选手：</span><strong style="color: blue;"><span style="color: black;">Meta AI 的MusicGen </span></strong><span style="color: black;">，它基于 Transformer 架构，靠上一段音频预测生成之后的音频片段。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">此刻</span> MusicGen 只<span style="color: black;">颁布</span>了 Demo ，能在 huggingface 上浅浅体验一波。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">例如</span>说生成一段嘻哈曲风的音乐，听起来很抓耳，节奏倒是蛮干净利落的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">和 Stable Audio 不太<span style="color: black;">同样</span>的是， </span><strong style="color: blue;"><span style="color: black;">MusiacGen 在生成音乐时，提示词会更自由一点，不仅有文字的选项，还<span style="color: black;">能够</span><span style="color: black;">弥补</span><span style="color: black;">有些</span>声音文件。</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">操作起来很简单，输入提示词，再把想参考的音乐片段直接拖到文件框内，<span style="color: black;">或</span>现场录音，当然音频提示<span style="color: black;">亦</span><span style="color: black;">能够</span>不填。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/faac5332df5d4b80a433301aa13a9fd6~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=OXrRWP7YGmSUgcfN%2BSEwFPYBTbw%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">虽然 MusiacGen 一次最长只能生成 30s 的音频，但有音频提示的加成，生成一段长音频<span style="color: black;">亦</span>不是不可能，<span style="color: black;">便是</span>会有点麻烦。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">只要每次生成 30s 的音频后，前后截取 10s <span style="color: black;">做为</span>之后的提示，最后拼接起来<span style="color: black;">便是</span>一段长音频了。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">不外</span>在<span style="color: black;">全部</span>体验过程中，有一点着实会劝退一大波人，那<span style="color: black;">便是</span>它生成的速度实在是太慢了，三四分钟还算好的，离谱的是有时等了好几分钟，结果<span style="color: black;">忽然</span>弹出个崩溃了的弹窗。。。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/3bbf4cc8fef844d3b72459390cd53fb5~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=OCBK1dD9y74ZmgsAT5vZhHJRpfk%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">今年年初，</span><strong style="color: blue;"><span style="color: black;">谷歌<span style="color: black;">亦</span>发布了音乐大模型 MusicLM </span></strong><span style="color: black;">，在现有的作曲 AI 中，谷歌的这个功能最多。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">除了最<span style="color: black;">基本</span>的文字生成音乐之外， MusicLM 还搞了<span style="color: black;">有些</span>其他<span style="color: black;">花招</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">例如</span>说故事模式</span></strong><span style="color: black;">，<span style="color: black;">能够</span>让它生成一段 1 分钟长的音乐： 0~15s 冥想、 16~30s 醒来、 31~45s 跑步、 46~60s 结束。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">生成的音频听起来确实还挺符合<span style="color: black;">需求</span>的，但就还是老毛病，乐器的声音<span style="color: black;">不足</span>清晰，各个段落之间的转换<span style="color: black;">亦</span>有点生硬。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">还有看图配乐的功能</span></strong><span style="color: black;">，给出一个经典的拿破仑骑马穿越阿尔卑斯山的图，再对<span style="color: black;">照片</span>进行<span style="color: black;">有些</span>描述， MusicLM 就能给生成 30s 的配乐。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/e6198343f2484c1fb809672afca96668~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=PN5iwXqVsmewPCxbfKLxWscFmtU%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">这次听起还真有点戏剧的感觉。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">MusicLM <span style="color: black;">一样</span><span style="color: black;">无</span>对外<span style="color: black;">颁布</span>，想要体验只能在 AI Test Kitchen 上排队获取内测资格。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/0e5daa7def504d609c095c29aca067b4~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=0OkTrCeYww2WoY3QTN36R5AQhCE%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">OpenAI 的 MuseNet ，在三年前就<span style="color: black;">已然</span>在官网<span style="color: black;">颁布</span>了。</span></strong></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/b4327b2b4e954433880e4622b01eba43~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=wEvLrZp6BMtznfDSbAIqdECOHuA%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">不外</span><span style="color: black;">近期</span>这几年倒是没怎么更新，还是基于和 GPT-2 <span style="color: black;">同样</span>的技术。</span><strong style="color: blue;"><span style="color: black;">并且 3 年过去了，这个 AI 还<span style="color: black;">无</span>对外开放<span style="color: black;">运用</span>。</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">但<span style="color: black;">瞧瞧</span>它官网对 MuseNet 的介绍以及给出的示例，估摸着出来<span style="color: black;">便是</span>吊打上面模型的存在。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">先不说生成音乐的质量，就光是时长就<span style="color: black;">已然</span>很顶了，最多<span style="color: black;">能够</span>生成 4 分钟的音乐。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">对比上面<span style="color: black;">说到</span>的几个模型，生成音乐的质感<span style="color: black;">亦</span>是分分钟秒杀</span></strong><span style="color: black;">，世超从官网下载了个示例，<span style="color: black;">大众</span><span style="color: black;">能够</span><span style="color: black;">一块</span>听听。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">不说是 AI 创作的，我还真会以为是那个音乐大师编的新曲子，有引入、有高潮，乐器的声音<span style="color: black;">亦</span>很清晰，再简单<span style="color: black;">调节</span>下<span style="color: black;">便是</span>个完整的音乐作品了。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">当然，有<span style="color: black;">这般</span>的效果除了有神经网络的功劳外，训练用的数据集<span style="color: black;">亦</span>是起到关键<span style="color: black;">功效</span>的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">OpenAI 统共用了数十万个 MIDI 文件训练 MuseNet ，下面这张图<span style="color: black;">便是</span>用到的部分数据集，从肖邦、巴赫、莫扎特到迈克 · 杰克逊、披头士、麦当娜，从古典到摇滚到流行，几乎<span style="color: black;">各样</span>风格的音乐都能在里面找到。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/784fe9cf672f4a3ab40a69313eb50157~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=Qrr5XrRJbaD42n0dNsmYNW%2Bnys4%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">不止国外，国内这几年 AI 音乐<span style="color: black;">亦</span>是发展得火热，去年华为<span style="color: black;">研发</span>者大会上，就<span style="color: black;">颁布</span>了一款音乐 AI ：Singer 模型，网易云面向音乐人推出了网易天音，</span><strong style="color: blue;"><span style="color: black;">作词、作曲、编曲直接都能靠 AI <span style="color: black;">处理</span>。</span></strong></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/cf5ef4e2b32040278013a0d610b37c3e~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=4MOVq%2FxE0R329LhrgBjRCz%2BbX6U%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">在前不久的 2023 世界人工智能大会上，腾讯多<span style="color: black;">媒介</span>实验室<span style="color: black;">亦</span>展示了自研的 AI 通用作曲框架 XMusic 。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/84f5ee3a07534bfab88e6d4590432668~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=gaHO3PMesoVVMsC7p9oXrgBDQ8Y%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">总的<span style="color: black;">来讲</span>，这几个 AI 作曲模型<span style="color: black;">亦</span>算是各有千秋，想要的音乐风格基本都能生成，<span style="color: black;">乃至</span>有时生成的音乐不仔细琢磨还真听不出来是 AI 生成的，用在<span style="color: black;">有些</span>短视频中<span style="color: black;">亦</span>是能妥妥地 “ 蒙混 ” 过去。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">但若要以一个专业人士来看的话，上面这些 AI 恐怕都或多或少有些缺点，最<span style="color: black;">显著</span>的<span style="color: black;">便是</span>上面<span style="color: black;">说到</span>的那几个 AI ，它们生成的音乐在乐器演奏上几乎都不太清晰。</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">并且，和 AI 作画<span style="color: black;">同样</span>， AI 音乐<span style="color: black;">亦</span>是版权问题的一大重灾区，<span style="color: black;">因为</span><span style="color: black;">关联</span>法律还跟不上 AI 发展的速度，时不时就有 AI 侵权的官司。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">例如</span>今年 1 月份，美国唱片业协会向政府提交了一份侵权报告，提醒<span style="color: black;">她们</span>要<span style="color: black;">注重</span> AI 音乐侵权的问题。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/a5ed513dda7e4484a9aa789997a3b6f6~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=%2FWsWjkkUAoROdf6TyggRFLnmQB0%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">就连 MusicLM 的<span style="color: black;">科研</span>人员<span style="color: black;">亦</span>亲口承认了侵权问题，在论文中写到会有盗用创意内容的潜在<span style="color: black;">危害</span>。</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">原由</span>是在<span style="color: black;">实验</span>这个模型的过程中，<span style="color: black;">发掘</span>它在生成的音乐里，大概有 1% 是直接从训练的数据集中照搬过来的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">亦</span>难怪<span style="color: black;">此刻</span>大多音乐 AI 模型要么干脆不对外试用，要么<span style="color: black;">仅有</span> demo <span style="color: black;">或</span>排队内测，就连对外开放的 Stable Audio <span style="color: black;">亦</span>是反复强调自己的数据集是经过 AudioSparx 授权的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">抛开版权问题不说，<span style="color: black;">日前</span> AI 在音乐这块的发展确实是令人咋舌，拥抱 AI 音乐<span style="color: black;">亦</span><span style="color: black;">已然</span>是行业内的大势所趋。</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">像专门<span style="color: black;">供给</span>轻音乐的 AI 音乐<span style="color: black;">机构</span> Endel ，<span style="color: black;">已然</span>先后得到了华纳、索尼等音乐巨头的投资， AI 音乐创作平台 Soundful <span style="color: black;">亦</span>拿到了环球音乐、迪士尼、微软的投资。</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/f6772a7165f6455db1b9be1f0366ef31~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1727615761&x-signature=tOuRClpcSFA5y6O9oTEnb9P2upE%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">当然，入局 AI 音乐是出于<span style="color: black;">商场</span>以及科技趋势的考量，在音乐性与艺术性上，<span style="color: black;">日前</span>的 AI 还是远不及人类创作者的，而这<span style="color: black;">亦</span>是<span style="color: black;">将来</span> AI 最应该优先<span style="color: black;">思虑</span>的。</span></span></p>
你的见解真是独到，让我受益匪浅。
页:
[1]