Skip to content

Transcription

Compare local Whisper and OpenAI transcription in Yap, including bundled runtimes, offline use, privacy, and model readiness.

Local transcription

Yap uses whisper.cpp on your computer. Release packages bundle the runtime (whisper-server, with whisper-cli fallback); you still download a model during setup. You do not need to install Homebrew Whisper just to use the packaged app.

Once the runtime and model are available, local transcription works offline and has no API usage fee. Startup or switching models can require warmup. Wait for readiness before recording.

Speed depends on model size, audio length, memory, and hardware. There is no fixed seconds-per-recording guarantee. Yap 0.5.0 uses Metal-first execution on Apple Silicon with CPU fallback. See release status.

Cloud transcription (OpenAI)

Select OpenAI in Settings, enter your API key, and save it. Yap sends audio to OpenAI for transcription. It needs internet access and an API account with available billing or quota.

Your key is stored in the local JSON configuration, not an encrypted credential store. Review OpenAI’s current API pricing and API data controls for costs and retention terms. Neither speed nor accuracy is guaranteed to exceed a local model for every recording.

QuestionLocalOpenAI
Where is audio processed?Your deviceOpenAI’s service
Works offline?After model downloadNo
Usage fee?NoneOpenAI API charges
Main performance factorsModel and hardwareNetwork and service latency
Local model needed?YesNo

Language and model choice

Choose .en models for English only. Choose a multilingual model without .en for other languages. The multilingual language-detection fix is included in 0.5.0. Read the model guide for selection and download steps.

Switching providers

Stop recording first, change the provider in Settings, and confirm the selected provider is ready. History keeps the original text; switching providers does not process old recordings again.