Apple's new SpeechAnalyzer API benchmarked against Whisper and its predecessor
Apple's new speech recognition engine outperforms competitors in accuracy.
1Apple's new SpeechAnalyzer API
With iOS 26 and macOS 26, Apple replaced the old SFSpeechRecognizer API with the new SpeechAnalyzer and SpeechTranscriber. The new API offers significantly higher accuracy. In LibriSpeech tests, it reduced the word error rate from 9.02% to 2.12% on clean speech. This means the new API is 3.5 to 4 times more accurate than its predecessor.
2Comparison with Whisper
Apple's new engine also outperformed OpenAI's Whisper Small model. It showed better performance on both clean and noisy parts of LibriSpeech while running approximately three times faster. Whisper's advantage lies in its support for more languages and its ability to run on non-Apple platforms.
3Accuracy and speed
SpeechAnalyzer is both more accurate and faster than Whisper Small. On an M2 Pro device, it can process an hour of audio in about 1.5 to 5 minutes. Whisper Small takes about three times longer for the same task. This is a significant advantage, especially for long audio recordings.
4Recommendations for developers
If your app still uses SFSpeechRecognizer, it is highly recommended to migrate to the new API. The new API makes far fewer errors, especially on long texts, and offers additional features like punctuation. For English, using SpeechAnalyzer instead of Whisper might be more sensible.
5Test methodology
Tests were conducted using the LibriSpeech dataset, which includes clean and noisy speech samples. All tests were run entirely on-device using the same code paths. This increases the reliability of the results. Additionally, the results are consistent with OpenAI's published Whisper results.
6User experience
SpeechAnalyzer offers a better user experience. It provides high accuracy even in noisy environments. Features like punctuation and capitalization make texts more readable. This is particularly important for long texts like meeting transcriptions.
7Conclusion and recommendation
Apple's new SpeechAnalyzer API is currently one of the best options for English speech recognition. Its optimization for Apple devices gives it an edge over competitors. Developers should consider migrating to this API to significantly improve user experience.