Added

• Parakeet Unified English. A second Parakeet model in Settings → Language & Model, for anyone who dictates only in English. It runs on the Neural Engine, writes its own punctuation and capitals, and on real dictations it got words right that Parakeet v3 missed, such as "create a PR" where v3 wrote "Kitab PR". It is English-only, so Parakeet v3 stays the choice for the other European languages; the picker marks Unified English Recommended when English is the only language you have selected, and nobody is switched over automatically. Your text always comes from one pass over the whole recording, which takes about a tenth of a second for a typical dictation. With live preview on, words also stream into the overlay as you speak, from a streaming model that downloads the first time you dictate with the preview on. The model is 614 MB; the streaming model is a further 591 MB.
• Use my dictionary with Parakeet (experimental, off by default). In Settings → Language & Model, Parakeet can now take your Dictionary into account, so rare names and jargon come out right more often. A small second model listens for your Dictionary terms and swaps them in where the audio supports it. That means it can occasionally replace an ordinary word with a similar-sounding Dictionary term, or drop a short word next to a name it fixed, and it adds about a third of a second to each dictation, which is why it stays off unless you turn it on. Plain words, short acronyms, phrases made of everyday words and the "Hey WrapScribe" wake-phrase hints are never used for this. It downloads a 103 MB helper model.
• Sound sets. Overlay & Sound has a new Sounds card with five sets of feedback sounds beside the macOS alert sounds: Click, Soft tone, Wood, Warm synth and Thock. Each set has its own Start, Stop, Done and Error sound, matched in loudness so the volume slider sounds the same whichever you pick, and ▶ lets you hear a set before choosing it. A set's Start and Stop also mark meeting recordings, and its Done plays when a transform lands; Thock has no Done sound. macOS sounds stay the default, with all your existing picks, until you choose a set, and your choice syncs to your other Macs.

Changed

• Parakeet v3 transcribes the whole dictation in one pass when you finish. The final text used to be whatever the live stream had settled on, and the stream stumbles over words where its windows meet: on real recordings it wrote "Parakid B2" for "Parakeet v2", "cyber changes" for "sidebar changes" and "key put in" for "keep it in", all of which a single pass over the recording gets right. That pass takes about a tenth of a second for a typical dictation and up to half a second for a long one; the live preview still streams as before.