[alan33d]

This PR adds WebIDL definitions and normative algorithm steps for opt-in speaker diarization to the Web Speech API (index.bs), as proposed in tracking issue #226 and companion explainer PR #227.

Summary of Changes

1. interface SpeechRecognition:
   • attribute SpeechRecognitionDiarization diarization; (defaults to "none")
   • attribute unsigned long minSpeakers; (defaults to 0)
   • attribute unsigned long maxSpeakers; (defaults to 0)
2. enum SpeechRecognitionDiarization:
   • Introduces "none", "speaker-turns", and "speaker-labels".
3. dictionary SpeechRecognitionOptions:
   • Adds SpeechRecognitionDiarization diarization = "none"; enabling feature detection and download gating via static SpeechRecognition.available() and install().
4. enum SpeechRecognitionErrorCode:
   • Adds "diarization-not-supported".
5. interface SpeechRecognitionResult:
   • readonly attribute boolean speakerChanged;
   • readonly attribute DOMString? speaker;
6. Algorithms & Normative Prose:
   • Parameter Validation: The start session algorithm throws a RangeError if minSpeakers > 0 and maxSpeakers > 0 and minSpeakers > maxSpeakers.
   • Capability Verification: Fires a SpeechRecognitionErrorEvent with error = "diarization-not-supported" if the requested diarization mode is not supported.
   • Boundary Enforcement: Requires that the recognition service segment results at speaker turns when diarization is "speaker-turns" or "speaker-labels".
   • Canonical Labeling: Sets speaker to 1-based index strings ("1", "2", ...) assigned in order of appearance within the session when diarization === "speaker-labels".
   • Immutability Contract: Guarantees that once result.isFinal is true, speaker and speakerChanged are immutable.

Related

• Tracking Issue: Add Speaker Diarization (Speaker Turn Attribution) to SpeechRecognition #226
• Explainer PR: Add explainer for speaker diarization proposal #227 (explainers/speaker-diarization.md)
• Related Issues: Add SpeechRecognitionResult Timestamps #191, Update SpeechRecognitionResult with speech timing attributes #192, What happens if the language isn't specified when using SpeechRecognition, and how to recognized multiple languages #208, The timestamp of various events is under-specified, and not implemented per spec in shipped implementations #213, Add explainer for speech recognition on AudioData #216

---

Preview | Diff