Apple has launched its new SpeechAnalyzer API, which provides significant improvements over the previous SFSpeechRecognizer, achieving up to 4x reduction in word error rates. This new API surpasses Whisper Small in both accuracy and speed when transcribing English speech on Apple devices, potentially influencing application developers to migrate to this updated technology.
The SpeechAnalyzer API by Apple has demonstrated superior on-device speech recognition, achieving the lowest word error rates compared to its predecessor, SFSpeechRecognizer, and Whisper models. Tests indicated a drop from 9.02% to 2.12% for clean speech and from 16.25% to 4.56% for noisy speech, highlighting its effectiveness in accurate transcriptions.
In performance tests, SpeechAnalyzer significantly outperformed Whisper Small on both clean and noisy speech tests, while consuming about one-third of the compute resources. Despite this, Whisper retains advantages in multilingual support and broader platform compatibility.
Developers are encouraged to transition from SFSpeechRecognizer to SpeechAnalyzer due to its clear superiority in transcription accuracy, especially for applications that require long-form speech recognition. The updated API not only produces more accurate results but also formats text in a more user-friendly manner.
As developers assess the capabilities of Apple's updated API versus Whisper models, the substantial improvements in SpeechAnalyzer might drive shifts in application development strategies, particularly on Apple’s ecosystem. Its performance could redefine standards for on-device speech recognition in English.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
Apple has launched its new SpeechAnalyzer API, which provides significant improvements over the previous SFSpeechRecognizer, achieving up to 4x reduction in word error rates. This new API surpasses Whisper Small in both accuracy and speed when transcribing English speech on Apple devices, potentially influencing application developers to migrate to this updated technology.