Where your recordings are transcribed
GoodParts turns your voice into text in the cloud. Your recording is uploaded, transcribed, and our copy is deleted. Onboarding tells you this before you record anything.
In the cloud
Your recording is uploaded and handed to our transcription provider. We then try to delete our uploaded copy from our storage straight away rather than keeping it while the transcript is produced; that delete is best-effort, and a failed one is reported to us rather than ignored. Provider-side cleanup of the temporary audio is best-effort after the transcript is ready. Cloud transcription is considerably faster for long entries. The provider transcribes through its EU processing endpoint, though its account metadata and logs may be processed in the US, and our account with it is opted out of its model-training program. The privacy policy names it; more on the your data page.
On this device
An on-device speech model — audio that never leaves the phone, and transcription with no signal at all — is the direction we want to go, and the app is built for two engines. It is not offered at launch: it needs a one-time model download of about 148 MB before the first entry can be transcribed, and we would rather ship that when it is reliable than make it your first experience of the app.
- Fast, even for long entries
- Our upload is deleted after handoff (best-effort)
- Provider account opted out of model training
- Nothing to download
- Not offered at launch
- Audio would stay on the phone
- Would work with no signal
- Needs a one-time model download
Analysis — the part that names your parts — always runs on our servers, whichever engine you choose. Only the transcript is sent, never the audio.
Changing your mind
Switching engines affects future entries only. Anything already transcribed stays exactly as it is.