Features

Transcription

How Shout converts speech to text — models, modes, and system audio capture.

How it works

Shout ships three speech engines and all of them run on your Mac. Parakeet (NVIDIA's FastConformer, through FluidAudio) is the default: English only, and the fastest. Whisper (through WhisperKit) covers 99 languages. Qwen3-ASR is the one to pick for Chinese, Japanese, Korean and several other Asian languages. Choose between them in Settings → Transcription — your audio never touches a server.

Note

100% local processing. Your audio never leaves your device. No internet connection is required for transcription.

Whisper models

With the Whisper engine selected, choose a model in Settings to balance speed and accuracy for your hardware. Quantized builds are smaller on disk and in memory.

ModelSizeSpeed

Tiny~40 MBFastest

Base~80 MBFast

Small~250 MBBalanced

Small (Quantized)~216 MBBalanced

Medium~800 MBSlow

Large Turbo~1.5 GBFast + accurate

Large Turbo (Quantized)~632 MBFast + accurate

Large~1.6 GBSlowest

Large (Quantized)~947 MBSlowest

Transcription modes

Live transcript

With the Parakeet engine, a Listening card streams your words while you speak. When you stop, the whole recording is transcribed again in one pass and that clean version is what gets pasted. Turn the card off in Settings → Output.

Retroactive Transcription

Go back in time using the timeline and transcribe any segment from the always-on buffer — even audio you didn't explicitly record.

Tip

The timeline shows a waveform of everything still in the buffer. Drag the selection handles to pick exactly the segment you want to transcribe.

System audio capture

Capture from other apps

Record audio from meetings, calls, podcasts, and videos alongside your microphone input. Enable system audio in Settings.

Careful

System audio capture requires the Screen & System Audio permission. Grant it in System Settings > Privacy & Security.