Apple introduced the new SpeechAnalyzer API at WWDC25 for on-device speech-to-text. A benchmark article compared its performance to OpenAI's Whisper and Apple's predecessor, showing competitive speed and accuracy. This API could disrupt existing ASR wrapper apps by providing a native, high-quality solution integrated into macOS/iOS, potentially making third-party transcription services less necessary. The benchmark tested on math lectures; SpeechAnalyzer was substantially faster than Whisper-Large-V2 and only slightly worse in accuracy. It runs entirely on-device, preserving user privacy.
Background
Automatic speech recognition (ASR) converts audio into text. Whisper is a popular open-source ASR model by OpenAI. Apple's previous speech API was less capable. The new SpeechAnalyzer API is part of Apple's push for on-device AI features.
References
Discussion
Commenters noted that better models like Nemotron and Voxtral exist, but agreed Apple's API could kill many Whisper wrapper apps. Some users reported successful integration into tools like Handy.computer, while others praised its speed for live transcription.