Apple's new transcription API 'SpeechAnalyzer' beats OpenAI's Whisper in speed tests
Apple's new speech-to-text API, SpeechAnalyzer , has been introduced in iOS 26 and macOS Tahoe , announced at Apple's annual developer conference WWDC25 . Finn Voorhees, editor of the Apple news site MacStories, tested SpeechAnalyzer and found that it was faster than OpenAI's transcription AI, Whisper .
Hands-On: How Apple's New Speech APIs Outpace Whisper for Lightning-Fast Transcription - MacStories
https://www.macstories.net/stories/hands-on-how-apples-new-speech-apis-outpace-whisper-for-lightning-fast-transcription/
Apple's New Transcription APIs Blow Past Whisper in Speed Tests - MacRumors
https://www.macrumors.com/2025/06/18/apple-transcription-api-faster-than-whisper/
Apple uses its own native speech framework for real-time transcription features in apps like Notes and Voice Memos, as well as for call transcription in iOS 18.1. To improve efficiency, iOS 26 and macOS Tahoe introduced the SpeechAnalyzer class and SpeechTranscriber module.
Voorhees' son and engineer Finn developed a CLI called ' Yap ' that can transcribe using the SpeechAnalyzer in macOS Tahoe. After hearing from Finn that 'SpeechAnalyzer is very fast,' Voorhees used Yap to convert the audio of a video file that was about 34 minutes long, had a resolution of 4K, and was about 7GB in file size, into text. As a result, the text conversion process was completed in just 45 seconds.
The following movie shows Yap actually transcribing the text.
According to Voorhees, the same file took 1 minute 41 seconds to transcribe using Whisper's Large-v3 model, which means that SpeechAnalyzer has reduced processing time by 55%. In addition, other Whisper-based tools, VidCap took 1 minute 55 seconds, and MacWhisper took 3 minutes 55 seconds using the Large-v2 model, so it is clear that transcription is extremely fast with the SpeechAnalyzer API. As for the quality of the transcribed text, Voorhees evaluated that there was 'no noticeable difference.'
Voorhees noted that the performance gains multiply exponentially when transcribing multiple videos or long-form content, and for those who regularly create subtitles or transcribe lectures, this increased efficiency could save hours of time.
The source code for Yap, developed by Voorhees, is available on GitHub under the CC0 1.0 Global License. At the time of writing, only the developer beta version of the required environment, macOS Tahoe, was released, and it is only available to developer users who are part of the Apple Developer Program .
GitHub - finnvoor/yap: 🗣 A CLI for on-device speech transcription using Speech.framework on macOS 26
https://github.com/finnvoor/yap/