Release Notes

The input device picked in Preferences was being ignored. Every recording came from whatever macOS had set as the system default, which is why dictation appeared to work only with the built-in microphone. Phantom Voice now records from the device you select.

Multi-channel inputs, which is most audio interfaces, could produce a completely silent recording. Phantom Voice now combines every channel of the incoming audio rather than assuming the microphone is on the first one, so an interface that puts the microphone on any channel still works.

And when a recording really does contain no sound, you are told which device was silent, instead of getting an empty file and no explanation.

Phantom Voice opened System Settings to the Microphone pane on every launch, even when permission had already been granted. It now asks only once, the first time, and never opens System Settings unless you ask it to.

Under Preferences, the trigger can now be set to press to start, press to stop instead of holding the key. Useful for long dictations, and for anyone who would rather not hold a key down. Holding remains the default.

See Trigger Modes.

The optional cleanup pass is a small local model, and a small model that meets a technical term it does not recognise can replace it with an ordinary word it does know. The sentence still reads correctly, which is exactly what makes it dangerous.

Cleanup is now checked rather than trusted. It may remove filler words and stutters, fix punctuation and capitalisation, and correct a mis-transcription into a term from your dictionary. If the result differs from what you said in any other way, it is discarded and your original transcript is used instead.

See Optional Cleanup.

The speech model accepts only a limited number of terms before it ignores the rest, and Phantom Voice used to send the whole list without saying so. Any term that does not fit is now marked not sent in Preferences.

See Personal Dictionary.