-
-
App Icon
-
Transcription View
-
Menu Bar Preview
Inspiration
The app was inspired by the Apple Watch Series 12's new Live Rewind feature, whereby the entire audio transcription lives on without the audio saved, preserving privacy, leaving only text.
What it does
Stenotypist does one thing and that is it starts transcribing on app launch, based on your previous settings (language, model type, audio source), in real time. It is non invasive, lives on your Mac's menu bar, and works locally using on device AI models, including Apple's SpeechTranscriber model as well as Alibaba's SenseVoice Small model as an alternative that supports transcribing in more than one language at once (English, Chinese, Japanese, Korean, Cantonese). Audio is fed through a temporary memory buffer, never saved on disk, for transcription purposes. Users can see phrases formed live, in the menu bar view, as the transcription process goes.
Transcripts may also be read later on, through the transcriptions window, and can be exported as markdown or text.
Users can use transcribe up to 3 sessions using Apple's model in English (US) language for free. Otherwise a subscription or one-time purchase is necessary for unrestricted sessions, additional languages and access to other models.
How I built it
It is built using SwiftUI, Apple's Speech framework, AVFoundation framework, RevenueCat for purchases, GRDB for transcript storage, SenseVoice, FunASR, Silero VAD and ONNX Runtime for the third-party model support. AI tools were also used to accelerate the building of the app.
Challenges we ran into
It turns out that even local transcription AI models can sometimes be quite hardware intensive. I settled only on 2 models (Apple's and SenseVoice) for now, as it was fast enough to run on any Apple Silicon Mac in real-time. I am planning to test other models and further improve on the app as time goes by.
Accomplishments that we're proud of
I think having a simple to use AI transcriber enhances workflows for certain groups of people, and to me that is an accomplishment on its own.
Requirements
The app requires macOS 26 and above. Apple Silicon Macs are required. Intel support is not available.
Log in or sign up for Devpost to join the conversation.