Apple's new SpeechAnalyzer API, benchmarked against Whisper and its predecessor (get-inscribe.com)

Apple has announced its new SpeechAnalyzer API, which dramatically outperforms its predecessor, SFSpeechRecognizer, and even the Whisper Small model from OpenAI in on-device speech recognition tasks. The SpeechAnalyzer achieved a word error rate (WER) of 2.12% on clean speech and 4.56% on noisy environments, showcasing a significant improvement over SFSpeechRecognizer's 9.02% and 16.25%, respectively. This new API runs approximately three times faster than Whisper Small, making it a highly efficient choice for real-time applications on Apple devices. The implications for the AI/ML community are substantial, as this development sets a new benchmark for on-device transcription, particularly for English, which was previously dominated by Whisper models. Developers are strongly encouraged to migrate to SpeechAnalyzer, as the accuracy alone justifies the switch. While Whisper maintains an advantage in multilingual support, SpeechAnalyzer is now the top choice for English transcription on iOS and macOS, highlighting Apple's continual advancements in making sophisticated AI tools more accessible directly through its operating systems.