|
← 判断力与美学 ← Judgment & Aesthetics
SAE 判断力与美学 · 余项之美
SAE Judgment & Aesthetics · Beauty of the Remainder
2026-05-21

AMYboard:定点运算的余项里长出了Juno-6的声音

AMYboard: The Sound of Juno-6 Growing from Fixed-Point Remainder

Han Qin (秦汉)

2026年5月18日,Brian Whitman和DAn Ellis发布了AMYboard——一块三十美元的10HP合成器电路板,可以装进Eurorack机架,也可以用USB供电摆在桌上。它运行MicroPython,可以在浏览器里编程,可以通过MIDI和CV连接外部设备。它身上插着一颗ESP32-S3芯片——成本大约三美元——和一个音频DAC。就这些。没有屏幕,没有操作系统,没有应用商店。你打开它,它就是一台合成器。你用Python告诉它要做什么,它就做。

但AMYboard只是冰山一角。真正的主角是AMY——一个用纯C语言编写的定点运算合成器引擎,大约一万行代码,已经跑了1883次提交,能在从树莓派到Playdate掌机到网页浏览器的几乎一切平台上运行。AMY能做Juno-6式的模拟合成、DX7式的FM合成、Karplus-Strong弦乐合成、PCM采样、波表合成、加法合成——120个复音声部,300多个预置音色,全都在一颗没有浮点运算单元的芯片上完成。这里是关键:AMY使用定点运算,不是浮点。ESP32-S3没有足够快的浮点硬件来实时合成音频,所以Whitman和Ellis被迫用整数模拟小数。这个技术选择不是出于美学,而是出于贫困。但它产生了一个意外后果:1983年的Roland Juno-6和Yamaha DX7也使用定点运算——因为那个年代的处理器同样买不起浮点。计算贫困的余项,跨越四十年后,在一颗三美元的芯片上与自己重逢。

这是一个典型的余项-已构循环。现代DAW(Ableton、Logic Pro)是已构:几千美元的软件跑在几千美元的电脑上,功能完备,范畴清晰,被命名为"数字音频工作站"。Eurorack模块化合成器也是已构:每个模块两三百美元,庞大的生态系统已经成型,被命名为"模块化合成"。Whitman和Ellis的凿子是极端简化——凿掉操作系统、凿掉图形界面、凿掉浮点运算、凿掉商业授权、凿掉一切不是"声音合成"的东西。留下的是什么?一个C语言库,一个Python REPL,一颗芯片。三十美元。这就是余项。但这个余项自身已经开始展现一种独特的美学:定点运算的量化噪声、固定精度的频率分辨率、整数溢出边缘的温暖失真——这些都不是设计出来的,它们是贫困的沉积物。

更深层的余项在于Tulip Creative Computer——AMY的姊妹项目,一台五十九美元的手持创意电脑。Tulip有一块七英寸触摸屏、MIDI接口、Wi-Fi,开机直接进入Python REPL。没有桌面,没有文件管理器,没有启动器。你面对的是一个光标。"创意电脑"这个范畴本身就是一个死去的余项——它的上一个化身是1985年的Amiga和Atari ST,在个人电脑通用化的大潮中被凿碎了。Tulip从这个死范畴的碎片中复活,但它不是怀旧:它有Wi-Fi、有WebAssembly端口(你可以在浏览器里运行同样的Tulip)、有一个叫TULIP~WORLD的内置BBS聊天室——Tulip用户可以从自己的设备上互相对话、分享代码。AMYboard的网页模拟器(AMYboard online)运行的是完全相同的C代码编译成的WebAssembly——硬件和浏览器产生一模一样的声音。三十美元的物理设备和零美元的浏览器标签页之间的区别,只剩下了触觉。

为什么现在看见它比以后看见它更重要?因为AMY/Tulip/AMYboard的命名缝隙还完全敞开。它不是商业合成器(太便宜、太开放)。不是DAW(太简陋)。不是Eurorack模块(太便宜、太可编程)。不是个人电脑(没有操作系统)。不是游戏机(虽然能跑游戏)。不是教育产品(虽然能教编程)。不是乐器(虽然能演奏)。Synthtopia报道它时叫它"DIY模块合成器";Hackaday叫它"MicroPython合成器工作站";Hackster.io叫它"Python驱动的音乐电脑"。每个命名都是局部的,没有一个能够覆盖它。这种命名的失败,正是余项还活着的证据。Whitman说他做Tulip做了好几年,经历了多次硬件迭代,希望有人觉得它跟他自己觉得的一样好玩。这种语气不是商业发布的语气,是一个人在自己家里造东西的语气。定点运算的余项正在生长,还没有人知道它最终会长成什么。

amyboard.com ↗

On May 18, 2026, Brian Whitman and DAn Ellis released the AMYboard — a thirty-dollar, 10HP synthesizer circuit board that fits into a Eurorack case or sits on a desk powered by USB. It runs MicroPython, can be programmed from a web browser, and connects to external gear via MIDI and CV. At its heart is an ESP32-S3 chip — roughly three dollars — and an audio DAC. That's it. No screen, no operating system, no app store. You turn it on, and it is a synthesizer. You tell it what to do in Python, and it does it.

But the AMYboard is only the surface. The real protagonist is AMY — a pure C, fixed-point music synthesis library of roughly ten thousand lines, with 1,883 commits, that runs on nearly everything: Raspberry Pi, Playdate handheld, Teensy, iOS, web browsers, the Electro-Smith Daisy, even Godot. AMY can do Juno-6 analog synthesis, DX7-style FM, Karplus-Strong strings, PCM sampling, wavetable, additive partials — 120 polyphonic voices, over 300 preset patches, all on a chip with no floating-point unit fast enough for real-time audio. Here is the crux: AMY uses fixed-point arithmetic, not floating-point. The ESP32-S3 cannot afford the computational cost of float operations at audio rates, so Whitman and Ellis were forced to simulate decimals with integers. This choice was not aesthetic — it was poverty. But it produced an accidental consequence: the 1983 Roland Juno-6 and the Yamaha DX7 also used fixed-point arithmetic, because processors of that era could not afford floating-point either. The remainder of computational poverty, separated by forty years, meets itself on a three-dollar chip.

This is a textbook chisel-construct cycle. The modern DAW (Ableton, Logic Pro) is construct: thousands of dollars of software running on thousands of dollars of hardware, feature-complete, cleanly categorized as "digital audio workstation." Eurorack modular synthesis is also construct: modules at two or three hundred dollars each, a mature ecosystem, named and digested as "modular synthesis." Whitman and Ellis's chisel is radical simplification — strip away the operating system, the graphical interface, the floating-point unit, the commercial license, everything that is not sound synthesis itself. What remains? A C library, a Python REPL, a chip. Thirty dollars. That is the remainder. And this remainder has begun developing its own aesthetic character: quantization noise from fixed-point arithmetic, frequency resolution limits from integer precision, warm distortion at the edges of integer overflow — none of this was designed. It is the sediment of poverty.

The deeper remainder lies in the Tulip Creative Computer — AMY's sibling project, a fifty-nine-dollar handheld creative computer. Tulip has a seven-inch touchscreen, MIDI ports, Wi-Fi, and boots directly into a Python REPL. No desktop, no file manager, no launcher. You face a blinking cursor. The category "creative computer" is itself a dead remainder — its last incarnation was the 1985 Amiga and Atari ST, shattered when the general-purpose PC won. Tulip resurrects from the rubble of this dead category, but it is not nostalgia: it has Wi-Fi, a WebAssembly port (you can run the identical Tulip in a browser), and a built-in BBS chat room called TULIP~WORLD where users talk to each other and share code from their devices. The AMYboard's web simulator (AMYboard online) runs the exact same C code compiled to WebAssembly — hardware and browser produce identical sound. The difference between the thirty-dollar physical device and the zero-dollar browser tab is reduced to touch alone.

Why does seeing this now matter more than later? Because the naming gap around AMY/Tulip/AMYboard is still wide open. It is not a commercial synthesizer (too cheap, too open). Not a DAW (too minimal). Not a Eurorack module (too cheap, too programmable). Not a personal computer (no operating system). Not a game console (though it runs games). Not an educational product (though it teaches programming). Not an instrument (though it plays music). Synthtopia calls it a "DIY modular synth"; Hackaday calls it a "MicroPython synth workstation"; Hackster.io calls it a "Python-powered music computer." Each name is partial; none can cover it. This failure of naming is evidence that the remainder is still alive. Whitman writes that he has been working on Tulip on and off for years, across many hardware iterations, and hopes someone finds it as fun as he has. That tone is not the tone of a product launch — it is the tone of someone building things in their own house. The fixed-point remainder is still growing, and no one yet knows what it will become.

amyboard.com ↗