A speech utterance that never starts also never ends, while speechSynthesis.speaking reads true

Fact

speechSynthesis.speak(u) flips speechSynthesis.speaking to true synchronously, before any audio exists. If the utterance can then never start — the page has not been interacted with and autoplay policy blocks audio, or the surface renders no sound — the browser neither starts it nor rejects it. onstart, onend and onerror all stay silent indefinitely.

Measured in a Chromium surface with 180 voices installed: 24 seconds, zero events, speaking === true and pending === false on every sample. The utterance is not queued and not failed; it is simply never spoken.

Why it matters

The idiomatic wrapper resolves a promise from onend and onerror, so that promise never settles and every await speak(...) deadlocks the state machine around it. The interface stays in its "speaking" state forever, with no error anywhere to explain it.

Both obvious health probes agree that nothing is wrong: speaking is true and pending is false, which is exactly what a healthy in-progress utterance looks like. So the stall reads as an application bug rather than a swallowed utterance.

How to apply

• Never settle a speak() promise on onend/onerror alone. Add a wall-clock watchdog sized from the text, e.g. 4000 + words * 620 / rate ms.
• Treat "no onstart within ~1.5 s" as the real signal that nothing is playing; speaking === true is not evidence of audio.
• Set a flag inside onstart and use that to decide whether the listener actually heard anything.
• The usual cause is the autoplay gesture requirement, so re-speak the first utterance on the first pointerdown/keydown rather than staying silent.