Player not loading? Watch on YouTube
This German-language tutorial follows the speaker's VoiceInk setup on Mac, with a comparison to Apple's dictation and subscription-based voice apps. The speaker describes VoiceInk as open source and available without a subscription, and also discusses an iOS app that they consider less mature.
The workflow has two stages: a speech recognition model transcribes recorded audio, then an optional language model cleans up the text before VoiceInk pastes it into the active field. Downloaded speech models support offline transcription. The demonstrated enhancement setup uses a provider API key, so that cleanup step sends the transcript to an external service.
The Mac walkthrough covers an Option-Space shortcut, downloading a speech model, and enabling enhancement. The speaker recommends disabling Remove Filler Words for German after demonstrating an unwanted word deletion. A correction to a meeting time illustrates how enhancement can remove an earlier, superseded instruction.
VoiceInk has preset prompts and a custom prompt editor. The speaker says their shared prompt works well with models of at least 30 billion parameters; this is advice for that prompt, rather than a general app requirement. Dictionary entries help with proper names, but the speaker says this feature requires enhancement. Word Replacement handles spoken emoji shortcuts. The tutorial also explains the distinction between local processing and cloud API costs.
