The result, up front Apple’s new SpeechAnalyzer is the most accurate on-device speech engine we tested. It beat every Whisper model we ship, including Whisper Small, on both the clean and the noisy half of LibriSpeech, while running roughly three times faster than Small. And the API it replaces, SFSpeechRecognizer, came last on clean speech: behind even Whisper Tiny, a 40MB model. Enginetest-clean WERtest-other WERModel size Apple SpeechAnalyzer (iOS/macOS 26)2.12%4.56%system Whisper Small (WhisperKit CoreML)3.74%7.95%~460MB Whisper Base5.42%12.51%~140MB Whisper Tiny7.88%17.04%~40MB Apple SFSpeechRecognizer (legacy)9.02%16.25%system Lower is better: WER is word error rate, the percentage of words an engine substitutes, drops, or invents. LibriSpeech test-clean is 2,620 utterances of…
Apple’s new SpeechAnalyzer API, benchmarked against Whisper and its predecessor
Advertisement