A demo that never errors still never ran
A WebGPU inference page can hand you four facts at once and none of them disagree: the UI renders, the GPU context comes back live, your click is accepted, and zero audio elements exist in the document. That was the state of my probe on November 4, against a WebGPU text-to-speech runtime advertising first audio in single-digit milliseconds, and the silence ran 75 seconds with no error event…
The WebGPU text-to-speech runtime claimed to deliver audio in single-digit milliseconds, but a test on November 4 showed 75 seconds of silence with no error event. The browser, Chrome version 156.0.8078.79, displayed no audio nodes or error nodes. The test claimed 26x real-time performance on an iPhone, but measured 75 seconds of silence instead.
The runtime checked through Apache-2.0 license, and the model weighed 43.7 MB after quantization to 8-bit integers. The demo was tested in Chrome on macOS and Windows, Safari on iOS, and Firefox on a Pixel 10 Pro.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
