Professional audio on the Web, what now?

<div class="title-layout">
  <div class="brand-lockup">
    <img src="assets/firefox-logo.png" alt="" />
    <img src="assets/firefox-wordmark.svg" alt="Firefox" />
  </div>
  <div class="title-copy">
    <h1>Professional audio on the Web, what now?</h1>
    <p class="speaker">Paul Adenot, Mozilla Corporation</p>
  </div>
  <p class="event-details">IEEE International Symposium on the Internet of Sounds
  <p class="event-details">29 October 2026</p>
  <p class="event-details">Cannes</p>
</div>

---

# Internet of sound?

---

## <https://internetofsounds2026.ieee-internetofsounds.org/call-papers>

---

<div class="irc-titlebar"><i></i><i></i><i></i><span>irssi</span></div>

<div class="irc-footer">
  <div class="irc-statusbar">[padenot(+i)] [#media]</div>
  <div class="irc-input irc-spacer"> </div>
</div>

<div class="irc-system">Time: 2014 (maybe?)</div>

<div class="irc-system">Location: irc.mozilla.org:6667 #media</div>

<div class="irc-line"><span class="irc-nick">person</span><span class="irc-msg">Hi did you know you support stereo in WebRTC?</span></div>

<div class="irc-line"><span class="irc-nick">person</span><span class="irc-msg">is it intentional?</span></div>

<div class="irc-line"><span class="irc-nick">person</span><span class="irc-msg">we rely on it to treat PTSD remotely, no other browser support this, it would be great it if didn't break</span></div>

---

<div class="irc-titlebar"><i></i><i></i><i></i><span>irssi</span></div>

<div class="irc-footer">
  <div class="irc-statusbar">[padenot(+i)] [#media]</div>
  <div class="irc-input"><span class="irc-typed">wtf</span><span class="irc-cursor"></span></div>
</div>

<div class="irc-system">Time: 2014 (maybe?)</div>

<div class="irc-system">Location: irc.mozilla.org:6667 #media</div>

<div class="irc-line"><span class="irc-nick">person</span><span class="irc-msg">Hi did you know you support stereo in WebRTC?</span></div>

<div class="irc-line"><span class="irc-nick">person</span><span class="irc-msg">is it intentional?</span></div>

<div class="irc-line"><span class="irc-nick">person</span><span class="irc-msg">we rely on it to treat PTSD remotely, no other browser support this, it would be great it if didn't break</span></div>

---

## Three lessons out of this

- You don't know who your users are, you never will, you can try, but **don't assume** ;
- Ease of access (a simple **URL**, a **web browser**, **any** device) is very powerful ;
- Standards, things not breaking are not _nice to have_, they are **foundational** and **enabling**.

---

<div class="section-layout">
  <span class="section-number">01 - Where we come from</span>
  <h2>Fifteen years of audio programming on the Web</h2>
</div>

---

# Who is this guy?

---

## Two ideas of what audio on the Web should be

- Humphrey & al., Firefox 4 (2011)
- Raw samples, in and out of JavaScript
- Low level
- You write the DSP
]
.col[
### Google: Web Audio API

- Chris Rogers, Chrome 10 (2011)
- A graph of native processing nodes
- High level
- You wire the DSP
]
]

Then a third: the MediaStream Processing API (Robert O'Callahan, Mozilla, 2011), DSP in workers.

---

> JavaScript is very much slower than heavily optimized C++ code and is not able to take advantage of SSE optimizations and multi-threading [...]. Optimized native code can be on the order of twenty times faster for processing FFTs as compared with JavaScript.

.attribution[Web Audio API, W3C First Public Working Draft, 15 December 2011, §15.4. Ed. Chris Rogers, Google.]

---

## So we standardized a graph of unit generators, implemented natively, because JavaScript was slow.

---

## Then worked our way down tirelessly from there

---

<div class="content-layout">
  <header>
    <h2 class="slide-heading">Fixing problems, one at a time, with standards</h2>
  </header>
  <div class="table-wrap timeline" style="margin-top: 0.5em">
    <table>
      <tbody>
        <tr><td>2013</td><td><strong>Web Audio API</strong>: all browsers supported.</td></tr>
        <tr><td>2013</td><td><strong>WebRTC interop</strong>: Firefox and Chrome can call each other.</td></tr>
        <tr><td>2013</td><td><strong>asm.js</strong>: JavaScript that compiles from native, runs fast. </td></tr>
        <tr><td>2017</td><td><strong>WebAssembly</strong> in every major browser. <strong>SharedArrayBuffer</strong> ships.</td></tr>
        <tr><td>Jan 2018</td><td><strong>Spectre</strong>: SharedArrayBuffer turned off everywhere overnight.</td></tr>
        <tr><td>2018–21</td><td><strong>AudioWorklet</strong>: Chrome, Firefox, Safari.</td></tr>
        <tr><td>2020–21</td><td><strong>SharedArrayBuffer</strong> is back, behind cross-origin isolation.</td></tr>
        <tr><td>2021</td><td><strong>Wasm SIMD</strong>. Web Audio 1.0 becomes a W3C Recommendation.</td></tr>
        <tr><td>2023–25</td><td><strong>WebCodecs</strong> and <strong>WebGPU</strong> in all engines.</td></tr>
        <tr><td>Now</td><td><strong>Document-Isolation-Policy</strong>, <strong>WebNN</strong>, <strong>Wasm ESM</strong> integration and <strong>components</strong>.</td></tr>
      </tbody>
    </table>
  </div>
</div>

---

<div class="section-layout">
  <span class="section-number">02 - Definitions</span>
  <h2>What does "pro audio" mean anyway?</h2>
  <p>What/who does the title of the talk even includes</p>
</div>

---

# That's the thing, it include most people

---

# Creator of systems<br>Users<span style="display: inline-block; width:0.95em;"></span>of systems

---

## Demos are cool <small style="font-size: 30px; font-weight: 300;">...the Web has never been short of that</small>

## Professional quality software is better

---

## Who are "the professionals"?

.chips[
- Musicians
- Producers
- Mixing & mastering engineers
- Sound designers
- Live sound
- Composers for film & games
]

.chips[
- Broadcast engineers
- Radio & TV producers
- Podcasters
- Post-production
- Archivists
- Restorers
- Forensic audio analysts
]

.chips[
- Acousticians
- MIR researchers
- Bioacousticians
- Speech scientists
- Teachers & students
- Sonification designers
]

---

<div class="section-layout">
  <span class="section-number">03 - Reputation</span>
  <h2>The Web has a bad rep</h2>
</div>

---

## "The Web is..."

- slow

- garbage collected, so it glitches

- high latency, not real time

- missing features I need

- not for mobile

- for cute demos, not for serious work

---

## How big is the Web slowdown?

The part of your code that you write runs as WebAssembly. Measured against native:

- **1.45–1.55×** on average, **2.1–2.5×** at worst, on SPEC CPU (Jangda et al., 2019)
- The 2011 spec said native was **20×** faster for FFTs

Let's take the worst reasonable case: **everything is twice as slow**.

.footnote[Jangda, Powers, Berger, Guha. "Not So Fast: Analyzing the Performance of WebAssembly vs. Native Code", USENIX ATC 2019. Firefox and Chrome, respectively.]

---

<div class="chart-layout">
  <header>
    <div>
      <p class="eyebrow">Meanwhile, hardware</p>
      <h2 class="slide-heading">Laptop CPUs, one thread</h2>
    </div>
    <div class="chart-steps">
      <button class="chart-step" data-step="halved">÷ 2</button>
      <button class="chart-step" data-step="apple">Apple Silicon</button>
    </div>
  </header>
  <div class="chart" data-chart="laptops" role="img" aria-label="Single-thread PassMark scores of 1,134 laptop CPUs from 2012 to 2026, with the yearly median rising from about 975 to 3,830."></div>
</div>

---

## Wasm on a good laptop of 2026 ≈ <br>Native on a good laptop in 2022

and we wrote good software in 2022 iirc

---

## And remember: the browser is native software

- The only tax you pay is WASM, for safety, but compile from native code
- Lock free real-time programming idioms are the same
- Network speed, latencies are the same
- Audio latencies are the same
- Disk IO latencies are the same
- Codecs are better and ready to use

---

## What's still missing, honestly

- Lossless, unprocessed audio in and out of WebRTC, end to end

- Multichannel devices: channel mapping and selection, consistently across browsers

- Device APIs (MIDI, USB, Serial, HID): not uniformly available

- Long-running background work on mobile

- Large project files: storage exists, directory access isn't universal

.footnote[See also: "Wasted Bandwidth: The Case for Lossless Audio in Networked Music Performance", Severi, this conference (MT 58).]

---

## Obligatory AI slide

.ml-grid[
.col[
- **Wasm SIMD + threads**: neural denoisers in calls today (Jitsi, Krisp, Zoom), Neural Amp Sim, etc.
- **WebGPU**: now in all browsers
- **WebNN**: higher level, faster tensor computing with xPUs. Running in Firefox, as of this week
- **Web Speech**: on-device real-time low-latency multinlingual hardware accelerated speech recognition
]
.figure[
<img src="assets/webnn-firefox.avif" alt="A MacBook running a local Firefox build: webcam images restyled as The Starry Night by WebNN" />

getUserMedia → WebNN style transfer, 100+ fps, Core ML, sandboxed and remoted. Local Firefox build (not shipped).
]
]

.footnote[This conference: "Neural Inference of Realtime Audio in the Web: Porting Anira to WebAssembly", von Raumer & Schulz (MT 60).]

---

## Is the web ready for your research and products?

yeah don't worry about it

- **Flexibility**: can we do everything we want? Mostly. The list of exceptions
is short, known and top people are on it

- **Performance**: mostly yeah, except if you're at the edge of the edge (but
then who are your users?)

- **Network**: come on... your pick of `fetch()`, P2P WebRTC, QUIC (UDP)

- **Real time**: yes, if you write real-time code, same as in native

- **Local AI Inference**: yes, and it's getting better fast

---

<div class="section-layout">
  <span class="section-number">04 - The promise</span>
  <h2>What the Web gives you that native can't</h2>
  <p class="section-note">Putting on my salesman hat</p>
</div>

---

## The Web is not the Internet.

The Internet of Sounds is larger than the Web. But every one of its systems needs a client, and there's one on every device on Earth.

---

## Distribution

- No installer

- No code signing, no notarization, no store review, no EULA

- Every OS: Windows, macOS, Linux, ChromeOS, Android, iOS and more

- Updates: reload

- Your colleague clicks a link at tries your stuff

- Your paper's reviewer clicks a link and runs your system

---

## It works today, it will work tomorrow

- A Web Audio demo from 2013 still runs, today, in every browser
- A website from 2004 renders the same
- 32 → 64 bit, `x86-64` → `aarch64`, new OS every year: some native code need
recompilation and fixes, Web pages didn't notice
- Your research artefact can outlive your PhD, your grant, and your lab's website

---

## Universal motto of people working in Web standards and on browsers:

## _Don't break the web_

---

## UI and accessibility, built in

- HTML and CSS: the most-used UI toolkit ever made, and the most debugged
- Accessibility by default: use the platform, screen readers just work
- Most native plugin GUIs are custom-drawn, opaque blob to a screen reader
- Resist the temptation to go full Canvas2D

---

## Networking, built in

- **WebRTC**: peer-to-peer, NAT traversal, etc., audio+video and data channels
- WebSockets, `fetch` streaming, WebSockets, WebTransport, from classical HTTP to TCP to spicy UDP
- **WebCodecs**: WebRTC media too opinionated? Reimplement it with WebCodecs and DataChannel, dropping down one level lower

.footnote[This conference: "Browser-Native Real-Time Multi-directional Streaming of Third-Order Ambisonics", Ferrari et al. (IWNIA 6). Sixteen channels, in a browser.]

---

## Portability, tested

- **web-platform-tests**: one test suite, all engines [wpt.fyi](https://wpt.fyi/results/webaudio)
- **Interop 202x**: browser vendors agree each year on what to fix so everybody can use it
- **Open source engines**: when it breaks, read why. LLMs make Gecko, Chromium and WebKit approachable. Send a testcase (even better if WPT), send a patch even.

<br>
## Pro audio APIs in interop next time?

---

## Security, privacy, trust

- Try to break a browser sandbox, **GOOD LUCK!**
- Extremely strong security model by default, cannot be bypassed
- Inherent trust
- We pay cash for exploits, and time from knowning the vuln to a new release is sometimes under 24h

---

## You don't think about security when you're clicking a link do you?

---

## Do you run random people `.exe` ?

---

## Are you persistent enough to try to run a python app untouched since 2014 ?

---

<div class="section-layout">
  <span class="section-number">05 - Reality check</span>
  <h2>Who actually uses it?</h2>
  <p class="section-note">Users, yes. Pros, it depends.</p>
</div>

---

<div class="content-layout">
  <header>
    <p class="eyebrow">Professional adoption in audio related fields</p>
    <h2 class="slide-heading">Uneven, to say the least</h2>
  </header>
  <div class="table-wrap" style="margin-top: 3em">
    <table>
      <thead>
        <tr><th>Field</th><th>Adoption</th></tr>
      </thead>
      <tbody>
        <tr><td>Calls, conferencing</td><td>Quasi-ubiquitous</td></tr>
        <tr><td>Broadcast & podcast</td><td>Real, professional, daily</td></tr>
        <tr><td>Research</td><td>Popular but far from the norm, <strong>should
              increase 👀</strong></td></tr>
        <tr><td>Education</td><td>Quite popular</td></tr>
        <tr><td>Art</td><td>Somewhat popular, not the norm</td></tr>
        <tr><td>Music production</td><td>Lots of users. How many records released?</td></tr>
        <tr><td>Film, post-production</td><td>A few professionals we know of</td></tr>
      </tbody>
    </table>
  </div>
</div>

---

## The biggest pro audio deployment nobody calls pro audio

- Hundreds of millions of people, daily, in a browser tab
- Echo cancellation, noise suppression, automatic gain control, any DSP via Web Audio
- Peer-to-peer (or relayed), multichannel, adaptive codecs
- sub-10ms audio roundtrip (depending on the hardware)
- And now neural denoisers, voice isolation, running in the page

.footnote[Jitsi Meet: RNNoise in Wasm, in an AudioWorklet (2022). Krisp JS SDK: Wasm in a worker + AudioWorklet. Zoom's Web client: its native media stack compiled to Wasm.]

and that's not even talking about the video part!

.footnote[Jitsi Meet: RNNoise in Wasm, in an AudioWorklet (2022). Krisp JS SDK: Wasm in a worker + AudioWorklet. Zoom's Web client: its native media stack compiled to Wasm.]

---

## This week, in this building

At least eleven contributions to IS2 2026 run in a browser:

.chips[
- Anira in WebAssembly (MT 60)
- Strummin' Gym (MT 75)
- Third-order Ambisonics streaming (IWNIA 6)
- Server-side binaural on the Web (IWNIA 82)
- openDAW (D7)
- WAM guitar platform (D57)
- WAM Jam Party, WebXR (D138)
- WAM Academy (D140)
- Cascade Delay (D153)
- Sonification webapps, Chafe (D166)
- Agentic composition in a browser (PSI 173)
]

and we should celebrate this: accessible, reproducible, runnable research is cool.

---

## but something is missing...

---

## Can I combine two systems presented today to make something novel?

---

<div class="section-layout">
  <span class="section-number">06 - The thesis</span>
  <h2>Why isn't adoption uniform?</h2>
  <p class="section-note">it's been 15 years, goddammit</p>
</div>

---

## The unit generator

"[...] a set of fairly universal building blocks and give the musician both the task and the freedom to put these together into his or her instruments."

---

## The unit generator wasn't a DSP idea. It was an interface idea.

A cool idea composable with another cool idea is better than a cool idea

---

<div class="content-layout">
  <header>
    <p class="eyebrow">1996: Steinberg VST</p>
    <h2 class="slide-heading">Success story: the audio plugin</h2>
    <p class="slide-subtitle">Then AU, AAX, LV2, CLAP. A flywheel that has been spinning for 30 years.</p>
  </header>
  <div class="signal-path">
    <div class="signal-step"><span>Developers</span><strong>ship plugins</strong></div>
    <div class="signal-step"><span>Hosts</span><strong>load any plugin</strong></div>
    <div class="signal-step"><span>Users</span><strong>learn, use, depend on, combine</strong></div>
    <div class="signal-step"><span>Money</span><strong>funds more plugins</strong></div>
  </div>
</div>

---

## Inertia. INERTIA!

A professional setup is years of presets, sessions, licences and muscle memory

---

## Platform vs. ecosystem

- Web Audio, AudioWorklet
- WebAssembly, SIMD, threads
- WebRTC, WebCodecs
- WebGPU, WebNN
- Web MIDI
]
.col[
### Ecosystem: too few people

- A plugin format
- Parameters, automation, presets, state
- Host ⇄ plugin negotiation
- Session interchange
- Discovery, licensing, a market
]
]

---

## The `AudioNode` is not a format.

We standardized low level blocks, not interchangeable units for an ecosystem

---

## People need to agree on standards

It's standards all the way up

---

## Another success story closer to "sounds" than "internet"

[Eurorack](https://doepfer.de/a100_man/a100m_e.htm): 1995, Germany, Dieter
Doepfer decides to specify a format (based on the
[Eurocard](https://en.wikipedia.org/wiki/Eurocard_(printed_circuit_board)
format):

- **a height**: DIN 41494 / IEC 297-3 / IEEE 1001.1 (19" rack, 6u)
- **a width**: the horizontal pitch, HP, 1/5"
- **a power supply specification**: +12V and -12V rail, and also +5V
- **ribbon cables** for power and some CV
- a spaghetti dish worth of **3.5mm jacks** (and jack sockets)
- unipolar or bipolar **CV**
- **1V / octave**

---

## Back to computers

Let's not pretend nothing has been started

---

## Web Audio Modules: the Web-native answer

- A plugin is a JSON descriptor and an ES module, loaded from a URL
- `WamNode` is an `AudioNode`, `WamProcessor` is an `AudioWorkletProcessor`
- Parameters, state save/restore, MIDI, automation, transport, OSC events between plugins
- The GUI is a DOM element, usually a Web Component (accessible!)
- Groups sandbox plugins from each other, hosts can nest
- A community index, Faust and others exports WAM directly

<br>
<strong>In browser engineering circles, we'd say this is "very webby"</strong>

---

## WebCLAP: bring the native ABI to the Web

- A CLAP plugin (the native C ABI) compiled to wasm32
- Exports `clap_entry`, a function table, `malloc`; shared memory; optional WASI
- GUI through CLAP's webview extension: the plugin UI is HTML
- The same binary loads in a browser (AudioWorklet host) or in a native DAW (wclap-bridge, Wasmtime)

<br>
<strong>I don't have a saying, but this is very native-y and attractive for existing vendors</strong>

---

<div class="content-layout">
  <header>
    <p class="eyebrow">Two cultures</p>
    <h2 class="slide-heading">Port the Web to plugins, or plugins to the Web?</h2>
  </header>
  <div class="table-wrap" style="margin-top: 1em; margin-bottom: 2em;">
    <table>
      <thead>
        <tr><th></th><th>Web Audio Modules</th><th>WebCLAP</th></tr>
      </thead>
      <tbody>
        <tr><td>Level</td><td>Web APIs: AudioNode, AudioWorklet, DOM</td><td>C ABI in linear memory</td></tr>
        <tr><td>Language</td><td>JavaScript / TypeScript, or Wasm inside</td><td>Anything that compiles CLAP to Wasm</td></tr>
        <tr><td>GUI</td><td>Web Component</td><td>HTML, via CLAP webview</td></tr>
        <tr><td>Runs in</td><td>Browsers</td><td>Browsers and native DAWs</td></tr>
        <tr><td>Best for</td><td>Web-first developers</td><td>The existing plugin industry</td></tr>
      </tbody>
    </table>
  </div>
</div>

---

## All well and good, but what about more than audio on the local machine

Sensors, network, etc. ?

---

# I don't know I'm just an audio standards guy!

---

<div class="section-layout">
  <span class="section-number">07 - What now?</span>
  <h2>Your turn</h2>
  <p class="section-note">Standards on top of our standards.</p>
</div>

---

> It used to be on us to provide a good platform. <br>
> It's on the community to write the next standards.

---

## Conversely: why would those use cases switch to the Web?

Many won't, and that's fine. The Web wins where distribution, collaboration,
networking, accessibility and longevity matter.

That's a fair description of the Internet of Sounds.

---

## Concretely

- Pick an interchange format: WAM, WebCLAP, a web component, a WASM Component maybe? Ship plugins in it, write the host.

- Publish your research as a URL. Make it still run in ten years, share it now.

- Measure: native vs. Web, same machine, same DSP, publish the numbers, tell us what matters, we improve.

- Tell us what's missing, with a test case. File bugs, read the source, send patches, it's all open, we're all happy to discuss.

- Come to the W3C Audio Working Group, Media WG, WebRTC WG, many more WG, the WebAssembly CG, the Bytecode Alliance.

---

## 1960: the unit generator at Bell Labs

## 2026: a runtime for it on every device on Earth.

---