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

A benchmark by Inscribe found Apple's new SpeechAnalyzer API outperformed OpenAI's Whisper Small model and the legacy SFSpeechRecognizer on accuracy and speed. The tests ran on an Apple M2 Pro using macOS 26.5.1 and the LibriSpeech dataset. SpeechAnalyzer achieved a word error rate of 2.12% on clean speech and 4.56% on noisy speech, compared to 9.02% and 16.25% for SFSpeechRecognizer. Whisper Small scored higher error rates than SpeechAnalyzer on both test sets. SpeechAnalyzer also ran roughly three times faster than Whisper Small per second of audio. Inscribe said the results prompted it to switch its Auto engine default to SpeechAnalyzer for supported languages. The company noted Whisper retains advantages in language coverage and cross-platform support. The benchmark used production code paths and published raw transcripts for verification. Limitations include English-only read speech testing on a single machine.