内容来源：https://www.theverge.com/ai-artificial-intelligence/974209/fenix-flexin-billboard-hot-100-rubberz-ai-slop

内容总结：

Fenix Flexin新歌被指AI生成：是音乐进化还是“人工智能口水歌”？

凭借热门单曲《Rubberz》飙升至公告牌百强单曲榜第58位，洛杉矶说唱组合Shoreline Mafia成员Fenix Flexin近期在个人事业发展上取得了突破。然而，这首歌曲走红的同时也深陷争议旋涡——大量乐迷和业内人士怀疑，这首歌的创作与制作过程高度依赖，甚至完全由人工智能完成。尽管Fenix本人否认了这些指控，但他并未做出有效澄清，反而在多个疑点上越描越黑。

从音乐风格来看，《Rubberz》是Fenix的一次彻底转型。这位以陷阱说唱风格著称的西海岸歌手，忽然转向80年代英国合成器流行乐，甚至模仿莫里西式的假英国腔演唱。但风格转型并不可怕，创作人求新求变本是常事，真正引发质疑的是录音室版本中大量难以解释的“AI痕迹”。

歌曲的录音质量成为首个突破口。仔细听辨可以发现，歌曲中存在多个人工智能生成音乐常见的明显缺陷：踩镲音色生硬、人声合唱部分如同低码率MP3、混响尾音突兀截断。播客制作人查理·哈丁指出，这首歌的伴奏中充满了不自然的数字伪影，“几乎可以让人瞬间想到从Napster下载的64kbps劣质音源”，这恰恰暴露了当前生成式AI模型多基于网络非授权低质MP3训练的通病。

歌词质量更是被指为“AI典型逻辑混乱”。整首歌押韵简单工整，却严重缺乏上下文的逻辑联系和叙事深度。诸如“刷卡堆筹码，看你沉船”这类前言不搭后语的句子反复出现，立意后便再未展开，极符合人工智能编词“重格式、轻表意”的特征。多位词曲作者和乐评人测试后认为，歌词在多个人工智能写作检测工具中均被标记为AI生成，甚至AI大模型Claude也给出了同样判断。

演唱细节和时间编排同样站不住脚。哈丁批评这首歌的“流动感极度缺失”，没有一位合格说唱歌手会以如此别扭的切分音强调节拍中所有错误的音节，“让整首歌的节奏充满刻意而生的怪诞感”。更为致命的是现场表现。Fenix在多次现场活动中要么对口型尴尬，要么试图真唱却明显力不从心——他的真实人声完全无法复刻录音室版本中的音域和口音，即使献声，背景音仍始终压过其原声，尽管DJ故意短暂消音来证明“真唱”，但效果适得其反。

面对质疑，Fenix辩称一切差异归于后期人声处理以及AutoTune等效果器。但多位专业人士纷纷反驳：AutoTune本质是修正音准的工具，它能修饰瑕疵，绝无可能拓展歌唱者的音域，更不可能凭空“生成”一个英式口音。

还有观点认为，正因为歌曲在艺术层面乏善可陈，反而可能恰恰是“真作”——毕竟，没有哪个用AI的创作者会选用如此拙劣的模型来生成如此糟糕的旋律。哈丁对此表达担忧：“如果这确实是某个低端AI音乐公司生成的成品，观众对该作品的追捧无疑令人堪忧。更高级的AI模型在音频瑕疵和节奏自然度上早已取得明显进步。”

目前，Fenix仅公布了自称是Pro Tools工程文件的截图，但这一点并不足以洗清嫌疑。分析普遍认为，除非他公开展示原始的录音音频和工程源文件，否则围绕《Rubberz》的AI争议几乎不可能终结。在技术深度介入创作的当下，如何界定“人机合作”的边界，显然正成为音乐行业一道无法回避的新考题。

中文翻译：

Fenix Flexin最为人所知的身份是洛杉矶说唱双人组Shoreline Mafia的成员。但他最近凭借单曲《Rubberz》在个人事业上取得了成功，这首歌攀升至公告牌百强单曲榜第58位。然而，几乎立刻就有关于这首歌来源的质疑声出现，许多人猜测这首歌大部分——甚至完全是——由AI生成。Fenix否认了这些指控，但几乎没有采取什么行动来消除这些疑虑。

这首公告牌百强热单是AI垃圾产物吗？

我们无法百分之百确定，但很可能如此。

这首公告牌百强热单是AI垃圾产物吗？

我们无法百分之百确定，但很可能如此。

部分怀疑当然源于这首歌是一次戏剧性的风格转向。Fenix此前主要涉足受陷阱音乐影响的西海岸说唱。而《Rubberz》则深受80年代英国合成器流行乐的影响。人声旋律明显受到Morrissey的启发，Fenix甚至模仿了一口假英国口音。但艺术家被允许成长和尝试新事物。仅凭这一点并不能证明什么。

也许更能说明问题的是录音本身的质感。仔细检查会发现许多在AI生成音乐中常见的伪影和异常。踩镲带有某种脆裂的质感，副歌人声听起来像低比特率的MP3，混响尾音被随意地突然切断。

在与The Verge的对话中，词曲作者兼Switched on Pop播客联合主持人Charlie Harding指出了这些破绽。“幽灵般的和声人声毫无预兆地忽隐忽现。鼓的混响充满了不自然的数字伪影，令人联想到从Napster下载的64kbps MP3。因为大多数生成式AI公司都用未经授权的版权材料的低质量MP3来训练模型，我们听到的是明显劣化的乐器声。”

将这首歌输入AI音乐检测器并不能找到确凿证据。我用了五种不同的工具来检测这首曲子，它们都说这首歌有20%到30%的概率是AI生成的或至少有AI辅助。但这类工具不一定非常可靠，而且其发展速度远慢于音乐制作模型本身。

然而，部分辅助材料确实触发了AI检测器的警报。这张单曲的封面艺术图以及一条庆祝其成功的帖子，都被多个AI图像检测器以超过97%的置信度标记为AI生成。同样地，我将歌词输入多个AI写作检测器和Gemini，它们一致认为歌词疑似由AI撰写。制作人兼YouTuber Curtiss King将歌词输入Claude，该工具也标记出多个迹象表明歌词疑似AI生成。

这些歌词正是AI经常输出的那种逻辑上近乎胡言乱语的内容。押韵简单而工整，不像是一个相当有成就的说唱歌手会写出来的东西。没有近似押韵或内韵。相反，一切遵循着基本的AA、BB格式，显然将押韵方案置于实际意义或叙事发展之上。

“刷着卡叠着筹码，我看见你击沉船只
留我在倾盆大雨中伫立，现在我买了一条沉重的钻石项链”

谁在击沉船只？我们是那艘被击沉的船吗？毫无头绪，这个意象再也没有被提及。同样地，副歌中的单句也许有意义，但组合在一起就毫无逻辑。

“赌桌上的大把钞票
看着本地有线电视
恨我能够做到
恨我能够做到”

先忽略“本地有线电视”这句——它很别扭（几乎让King陷入暴怒）——Fenix能做什么？为什么他恨自己能做到？这些歌词提出的问题比回答的还多。

这些歌词的演唱方式也很奇怪。第三段歌词似乎在“不是珠宝”这句上戛然而止，像是话说到一半。Harding指出它“完全没有流动感”。他说：“任何一个稍有能力的说唱歌手都不会用如此笨拙的方式唱出标题那句‘我就让橡皮筋弹回去’。切分音极其反常，重音全落在错误的音节上。”他接着解释说：“把这么多音节挤在一起，迫使‘the’这个词落在强拍上，而不是‘rubber’这个词，导致后续所有音节都以不自然的切分方式被强调。”

《Rubberz》的现场“表演”也没有给Fenix加分。最初几次录音室亮相中，这位说唱歌手都在尴尬地对口型。有时他似乎被这种别扭的流畅感和歌词绊住，仿佛他记不住自己的歌。老实说，他甚至没有很努力地去让任何人相信他真的在唱。

当Fenix真的开口唱时，情况只会更糟。多个片段中极为明显地看出，他无法还原录音中的音域和口音（而且他也不知道歌词）。在许多现场表演片段中，即使他在唱，播放的也是原声人声覆盖在他的声音之上。他的真实人声在混音中被压得极低，DJ会策略性地短暂切掉伴奏，试图证明歌曲的真实性，但失败了。

Fenix为自己辩护称，录音版和真人表演之间的差异归因于声音处理，包括AutoTune。然而，包括Harding、King和Anthony Fantano在内的多人都反驳了这一说法。King将AutoTune比作拼写检查，这个类比并不差。虽然它确实可以被推向极限以创造独特的机器人质感，但它的设计初衷是修正错误——它不会扩展表演者的音域或带来口音，正如Fantano所指出的。

也许最能证明《Rubberz》是真人创作的有力论据是它实在太糟糕了。Harding说：“我会认为这首歌是由一个从未用过80年代儿童练习键盘写过歌的人写的。”如果这首歌确实是用生成式AI创作的，那么显然是用了一个劣质模型。“我担心的是，当这首歌是由一家模型很差的三流生成式AI音乐公司创作时，观众竟然会觉得它有趣。更先进的模型已经在可听伪影和尴尬的演唱节奏方面做出了重大改进。”

Fenix发布了他声称是Pro Tools工程文件的片段。但这实际上证明不了什么。除非他公开发布原始录音和工程文件，否则他很可能永远无法摆脱AI的指控。而我看不到这在短期内会发生的可能。

更多内容：AI“音乐”领域的所有最新动态

热门文章

• 你的手机拍照本可以好看得多
• 禁止扫地机器人不会让它们更安全，只会更糟
• 有了Switch 2、iPhone和笔记本电脑的花样，Sharge Disk Pro 2终于成为一款值得随身携带的EDC产品
• 这款纹身是永久性的、无痛的，而且可能很快就会通过邮件送达
• 美国政府刚刚禁售了Roomba

英文来源：

Fenix Flexin is best known as a member of Shoreline Mafia, a rap duo from Los Angeles. But he’s recently found solo success with the track “Rubberz,” which has climbed to number 58 on the Billboard Hot 100. Almost immediately, though, questions were raised about the song’s origins, with many speculating that it was largely, if not entirely, generated by AI. Fenix has denied those accusations, but done little to dispel them.
Is this Billboard Hot 100 hit AI slop?
We can’t be 100 percent sure, but probably.
Is this Billboard Hot 100 hit AI slop?
We can’t be 100 percent sure, but probably.
Part of the suspicion is certainly due to the fact that the song is a dramatic left turn. Fenix has primarily trafficked in trap-influenced West Coast rap. “Rubberz,” on the other hand, is heavily influenced by ‘80s UK synth pop. The vocal melody is clearly influenced by Morrissey, with Fenix even adopting a fake British accent. But artists are allowed to evolve and experiment. That alone doesn’t prove anything.
Perhaps more telling is the simple quality of the recording. Close inspection reveals a number of artifacts and anomalies that are commonly found in AI-generated music. Hi-hats have a certain brittle quality, the chorus vocals sound like a low-bitrate MP3, and reverb tails haphazardly cut off.
Speaking with The Verge, songwriter and co-host of the Switched on Pop podcast, Charlie Harding, highlighted the giveaways. “Ghostly backing vocals duck in and out unexpectedly. The drum reverb is full of unnatural digital artifacting reminiscent of a 64kbps MP3 downloaded off of Napster. Because most generative AI companies have trained their model on low-quality MP3s of unlicensed copyrighted material, we hear audibly degraded instrumentation.”
Feeding the song into an AI music detector doesn’t deliver a smoking gun. I ran the track through five different tools, and they all said that there was between 20 and 30 percent chance that it was AI or at least AI-assisted. But those sorts of tools are not necessarily super reliable and advance much more slowly than music-making models themselves.
Some of the supporting materials, however, have tripped up AI detectors. The single’s cover art and a post celebrating its success were both flagged as being AI-generated with over 97 percent certainty by multiple AI image detectors. Similarly, I fed the lyrics into multiple AI writing detectors and Gemini, which agreed they appeared to be written by AI. Producer and YouTuber Curtiss King dropped the lyrics into Claude, and it also flagged multiple signs that the lyrics appeared to be AI-generated.
The lyrics are the sort of logical near-gibberish that AI often spits out. The rhymes are simple and clean in a way that you wouldn’t expect from someone who is a reasonably accomplished rapper. There are no near rhymes or internal rhymes. Instead, everything sticks to a basic AA, BB format that clearly prioritizes the rhyme scheme over actual meaning or narrative development.
Swipin’ cards and stackin’ chips, I saw you sinkin’ ships
Left me standin’ in the pourin’ rain, now I bought a heavy diamond chain
Who is sinking ships? Are we the sinking ships? No clue, this idea never comes back up. Similarly, the individual lines of the chorus might make sense, but when combined are meaningless.
Bankrolls on the table
Watchin’ local cable
Hatin’ that I’m able
Hatin’ that I’m able
Ignoring the “local cable” bit, which is just awkward (and nearly sent King into a rage spiral), what is Fenix able to do? Why does he hate it? The lyrics raise more questions than they answer.
The way the lines are delivered is also odd. Verse three seemingly ends mid-thought on the line, “Not the jewelry.” Harding noted that it has a “complete lack of flow.” He says that, “No half-competent rapper would ever sing the title line ‘I just let the rubber band snap’ in such an awkward manner. The syncopation is highly unusual, emphasizing all the wrong syllables.” He goes on to explain that, “by smashing so many syllables together, it forces the word the onto the downbeat, instead of the word rubber, forcing all of the subsequent syllables to accent with unnatural syncopation.”
Live “performances” of “Rubberz” haven’t done Fenix any favors. The first couple of in-studio appearances featured the rapper awkwardly lip-syncing. At times it seems like the awkwardness of the flow and the lyrics trip him up, as if he can’t remember his own song. Honestly, he’s not even trying that hard to convince anyone that he’s actually singing.
Things only get worse when Fenix actually does sing. It’s abundantly clear in multiple clips that he is incapable of recreating the vocal range and accent from the recording (and that he doesn’t know the lyrics). In many of the clips of live performances, even when he’s singing, they’re simply playing the original vocals over his. His real vocals are kept extremely low in the mix, and the DJ strategically drops out the backing track briefly in an effort to prove the authenticity of the song, but fails.
Fenix has defended himself, saying the difference between the recorded version and any in-the-flesh performance is down to vocal processing, including AutoTune. However, multiple people, including Harding, King, and Anthony Fantano, have disputed this assertion. King compares AutoTune to spellcheck, and that’s not a bad analogy. While yes, it can be pushed to its limits to create unique robotic textures, it’s designed to clean up errors — it would not extend a performer’s vocal range or introduce an accent, as Fantano points out.
Perhaps the most compelling argument for “Rubberz” being authentic is just how bad it is. Harding says, “I would assume it was written by someone who has never written a song using an ‘80s kids practice keyboard.” If the song was indeed created using generative AI, it was clearly made with an inferior model. “I am concerned that audiences find this song interesting when it is created by one of the lesser generative AI music companies with a terrible model. More sophisticated models have already made big improvements in the audible artifacts and awkwardly sung rhythms.”
Fenix has posted clips of what he says are Pro Tools session files. But that doesn’t actually prove anything. Unless he shares the original recording and project files publicly, it’s likely he’ll never shake the AI allegations. And I don’t see that happening anytime soon.
More in: All the latest in AI ‘music’
Most Popular

• You could be taking way better photos on your phone
• The ban on robot vacuums won’t make them safer, only worse
• With Switch 2, iPhone, and laptop tricks, the Sharge Disk Pro 2 is finally a worthy EDC
• This tattoo is permanent, pain-free, and might soon come in the mail
• The US government just banned Roombas