Apple's New SpeechAnalyzer API Benchmarked Against Whisper and Its Predecessor
Apple's new speech recognition engine outperforms Whisper in both accuracy and speed.
1Apple's New SpeechAnalyzer API
Apple's new SpeechAnalyzer API shows significant improvement over the previous SFSpeechRecognizer. It achieved the lowest word error rate (WER) in both clean and noisy speech tests on LibriSpeech, offering developers a more accurate and reliable speech recognition solution.
2Comparison with Whisper
Apple's new engine outperformed OpenAI's Whisper Small model in both accuracy and speed. It runs approximately three times faster than Whisper Small while maintaining a clear advantage in word error rate, making it the best option for Apple hardware.
3Legacy API Performance
Apple's old SFSpeechRecognizer API showed the lowest performance in tests, even lagging behind small models like Whisper Tiny. The new API reduced the word error rate by 3.5 to 4 times, marking a significant improvement.
4Recommendation for Developers
If your app still uses SFSpeechRecognizer, migrating to SpeechAnalyzer is highly recommended. The new API not only improves accuracy but also delivers better results in details like punctuation and capitalization.
5Whisper's Advantages
Whisper still offers two key advantages: support for over 30 languages and compatibility with non-Apple platforms. However, for English speech recognition on Apple devices, SpeechAnalyzer is now the better choice.
6Speed Comparison
All engines ran faster than real-time. SpeechAnalyzer was three times faster than Whisper Small while producing more accurate results, a significant advantage for processing long audio recordings.
7Conclusion and Implications
Apple's new SpeechAnalyzer API is a significant step forward in speech recognition technology. It offers developers a more accurate and faster option while encouraging the evolution of open-source solutions like Whisper. This competition ultimately benefits users with better products.