Apple Boosts On-Device Multimodal Capabilities with Siri AI and SpeechAnalyzer in iOS 27 Public Beta
멀티모달 | Tue Jul 14 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 3 sources
Apple unveiled a fully revamped Siri AI in the iOS 27 public beta and introduced a new on-device speech recognition API that delivers a major leap in accuracy.
Analysis
[Apple] unveiled Siri AI in iOS 27 public beta [1][3]
- distributed as an opt-in beta program
- enables natural-language task instructions without opening apps
- integrates information across apps such as web
- and calendar
[Siri AI] integrated as a chatbot-style app across the iPhone [3]
- added a dedicated app for managing conversation logs
- combined with search to be callable from anywhere
- conversation retention selectable among 30 days
- 1 year
- or permanent
[Apple SpeechAnalyzer] launched a new on-device speech recognition API [2]
- replaces SFSpeechRecognizer starting with iOS 26/macOS 26
- recorded 2.12% WER on LibriSpeech test-clean
- outputs text with punctuation and capitalization
[SpeechAnalyzer vs Whisper] outperformed Whisper Small in on-device English speech recognition benchmarks [2]
- test-clean WER 2.12% vs Whisper Small 3.74%
- test-other WER 4.56% vs Whisper Small 7.95%
- approximately 3x faster processing speed than Whisper Small
[Inscribe] switched its default engine to SpeechAnalyzer based on benchmark results [2]
- prioritizes SpeechAnalyzer for supported languages
- validated comparisons across 5 engines using the same code path
- recommends migration away from continued SFSpeechRecognizer use