I'm working on a macOS transcription utility that uses Apple's Speech framework (SpeechAnalyzer) for speech-to-text. This is for meetings/interviews/podcasts where speaker identification is critical. The current limitation is speaker attribution — I need to identify which speaker is producing each segment of transcribed text. I have three questions:
- Native diarization in iOS 27 / macOS Golden Gate
Is native diarization coming in the fall release? I've reviewed the WWDC 2026 session catalog and found no mention of diarization in SpeechAnalyzer or elsewhere. I'm probably going to use FluidAudio for speaker attribution, but I'd strongly prefer a native solution if one exists or is planned. Do I need to stay with third-party libraries, or is this coming?
- Core AI and custom models
The new Core AI framework was announced for on-device model deployment. Can I train or integrate a custom diarization model via Core AI? If yes, are there sample implementations or documentation for audio-processing models?
- Core Audio framework updates
Were there any Core Audio API-level additions announced at WWDC 2026 that might support audio analysis or speaker detection downstream? I saw no dedicated session, but wanted to verify.
Thanks for any guidance on this.