July 13, 2026
Speech Wars: Apple Strikes Back
Apple's new SpeechAnalyzer API, benchmarked against Whisper and its predecessor
Apple’s new voice tool just embarrassed its old one — and the comments got messy
TLDR: A new test says Apple’s latest built-in speech-to-text system is far better than its old one and even beats popular Whisper models on Apple hardware. Commenters split instantly between impressed beta-watchers and skeptics shouting that real-world voice assistants still feel slow and messy.
Apple dropped a new built-in speech-to-text tool, SpeechAnalyzer, and according to this first big public test, it didn’t just win — it steamrolled the old Apple system and even beat several versions of Whisper, the fan-favorite transcription tool. In plain English: on newer Apple devices, it turned spoken words into text more accurately and much faster, which is a very big deal for anyone who records meetings, notes, or interviews. The benchmark maker says the old Apple option now looks almost painfully outdated, with around four times as many mistakes on the same audio.
But the real popcorn-worthy action was in the community reaction. Some readers were impressed and instantly turned into unpaid product managers, begging for a rerun on the iOS 27 beta to see whether Apple’s next update pushes things even further. Others were absolutely not ready to hand Apple the crown. One especially fiery commenter declared “Apple’s isn’t even usable” and then doubled down by calling downvotes from “Apple fanboys” a badge of honor — which is basically comment-section gasoline. Meanwhile, another thread veered into classic Apple drama: if this is so smart, why does Siri still feel slow on older phones compared with talking to ChatGPT? That mismatch became the sneaky subplot here. So yes, Apple may have won the benchmark, but in the comments, the war is still very much on.
Key Points
- •Inscribe benchmarked Apple SpeechAnalyzer, Apple SFSpeechRecognizer, and three Whisper models on 5,559 LibriSpeech utterances using identical on-device test conditions on an Apple M2 Pro.
- •Apple SpeechAnalyzer achieved the lowest reported word error rates: 2.12% on LibriSpeech test-clean and 4.56% on test-other.
- •Apple’s legacy SFSpeechRecognizer performed worst on clean speech at 9.02% WER, behind even Whisper Tiny at 7.88%.
- •The article says migrating from SFSpeechRecognizer to SpeechAnalyzer reduces word error rate by roughly 3.5x to 4x and improves output formatting with punctuation and casing.
- •Inscribe says SpeechAnalyzer beat Whisper Small on both benchmark splits while running at roughly one-third of Whisper Small’s compute time per second of audio, leading the product to prefer SpeechAnalyzer for supported languages.