Suno and the 'air fryer anthem': how neural networks turn spectrograms into music

⚡ The gist in 5 seconds

• DJ NARODNII's track 'You Are My…' racked up over 150K videos in a week and earned the nickname 'the air fryer anthem'.
• The author intended it as a parody of formulaic AI music — some listeners took it at face value.
• Context: in February 2026 Suno reported 2 million paid subscribers and $300 million in annual revenue, and in July it built track generation into iMessage.

🔍 What happened

DJ NARODNII released 'You Are My…' — and within a week, more than 150K videos were made to it. The author himself called the song a jab at low-quality AI music and its listeners. The audience split: some believed it, others saw meta-irony and a diss aimed at tasteless neural tracks. The author agreed with that interpretation in the comments. Then came complaints to streaming services — claiming the lyrics romanticize violence and demean women. The author's response: fictional imagery doesn't carry over into real life. A detailed breakdown was published by the SpeShu.AI blog on Habr.

The mechanics are simple if you break them down step by step. To a computer, sound is a stream of numbers: CD quality means 44,100 measurements per second per channel, and a minute of stereo is over 5 million numbers. Working with such a stream directly is inconvenient. So the recording is cut into short overlapping windows and a spectrogram is computed: time runs horizontally, frequency vertically, and brightness represents loudness. Frequencies are converted to the mel scale. It mimics human hearing: we can hear the difference between 200 and 300 Hz, but barely between 10,000 and 10,100 Hz. Lows are stretched out, highs are compressed, and the model's attention goes where the human ear goes.

That's how OpenAI's Whisper works (September 2022): 680,000 hours of audio with transcripts, resampling to 16 kHz, an 80-channel log-mel spectrogram, 25 ms windows with a 10 ms step. A transformer then reads this image almost like text — and learns to predict the next frequency the same way a language model predicts the next word.

🧠 Our take

The funny thing is different: the author aimed at the formula of a neural pop song and hit it dead center, without a miss. A saccharine melody, polished vocals, simple lyrics — the recipe worked to the tune of 150K videos. According to Yota, in 2026 traffic to music neural networks grew by more than 20%, and time spent in them by 45% year over year. Our assessment: the parody doesn't break the formula — it confirms it. The 'air fryer' meme works as a label here: the audience laughs at everyday life and, at the same time, at itself.

What this changes in practice. Suno entered February 2026 with 2 million paid subscribers and $300 million in annual recurring revenue. In July of the same year, song generation moved into iMessage: a text or voice message in a chat turns into a track. The barrier to entry drops to zero — no separate app needed anymore. We believe that from here on, the competition won't be over track quality but over speed: whoever delivers a song straight into a conversation first wins.

Who wins? Platforms and meme creators — they get the reach and the subscriptions. Who loses? Most likely, streaming services: complaints about lyrics are already coming in, which means they'll have to build moderation for content generated thousands of times a day. And the third camp — musicians, for whom being better is no longer enough: they have to be more visible in the feed.

⚠️ What we don't know

The Yota and Suno figures come from the article, with no primary sources cited — treat them as a reference point, not audited reporting. There's no breakdown of the 150K videos by platform or time period. How much money the track made, how many complaints actually reached the services, and how the reviews ended — unknown. The piece's weak spot is that it mixes hype and technology: the former is easy to verify, the latter only via model documentation.

Forecast: within a quarter, at least one major streaming service will add a label or filter for AI music. The condition — the wave of complaints and memes doesn't fizzle out before then.

Source: Habr / SpeShu.AI