Apple's new SpeechAnalyzer API, benchmarked against Whisper and its predecessor

	Apple's new SpeechAnalyzer API, benchmarked against Whisper and its predecessor(get-inscribe.com)
	543 points by get-inscribe 20 hours ago \| 219 comments
	tl;dr: Apple's new SpeechAnalyzer API (iOS/macOS 26) beat Whisper Small and all smaller Whisper variants on LibriSpeech, achieving 2.12% WER on clean speech and 4.56% on noisy speech while running ~3x faster than Whisper Small on an M2 Pro. The legacy SFSpeechRecognizer came in last at 9.02%/16.25% WER, making migration a clear win. Whisper still holds advantages in language coverage (~30 locales for SpeechTranscriber vs. 100+) and cross-platform support, but for English on Apple hardware, the built-in engine is now the strongest on-device option.
	HN Discussion: ↓Whisper small is outdated; benchmark should include newer SOTA models like Parakeet, Voxtral, or Nemotron ↑Personal testing confirms SpeechAnalyzer is fast and accurate for real-world use cases ~SpeechAnalyzer's streaming/live transcription capability is a major UX advantage not captured by WER benchmarks ↓Third-party tools like Willow or Voxtral outperform both Whisper and SpeechAnalyzer for specific use cases •Neutral additions noting implementation details, storage behavior, or integration efforts