Interview recordings are the most sensitive audio most researchers handle, and a cloud transcriber is an upload you have to explain to an ethics board. templobe runs an English speech model on your Mac instead: the recording never leaves the machine, and a full round of interviews can be queued as one batch.
mp3, wav, m4a. Up to 50 files per batch. No upload. English only in this beta.
| What matters | templobe | Typical cloud service |
|---|---|---|
| Where recordings go | Stay on your Mac | Uploaded to a vendor's servers |
| Third parties in the chain | None for audio | The transcription vendor, and any subprocessors |
| Consent form implications | No audio-processing vendor to disclose | Usually requires a data-processing disclosure |
| Deleting the recordings | Delete local files; no remote copy | Depends on the vendor's retention window |
| Cost per project | Free for beta users | Per-minute, scaling with total interview hours |
| Works offline | Yes, after one sign-in | No |
| Batch of interviews | Up to 50 files per batch | Often an upload-per-file workflow |
| Speaker labels | Not in this beta | Offered by some providers |
This compares the common cloud model, not one vendor. Check your institution's rules and each provider's data-processing terms before sending interview audio anywhere.
Move the mp3, wav or m4a files off the recorder onto the Mac. Up to 50 per batch, 2 GB per file, 5 GB total.
The batch runs against a speech model installed on the machine. No file is uploaded, and the Mac can stay offline while it works.
Check each transcript against playback, then export txt, md, srt or vtt. Speaker labels are not generated in this beta, so speaker turns are marked by hand.
No. Speech recognition runs on your Mac with a local model. There is no upload step for audio.
Yes, one account to activate the app and download the model. That account never receives your recordings, and transcription keeps working offline afterwards.
No. Speaker diarization is not part of this beta, so interviewer and participant turns are not separated automatically. The transcript keeps the speech in order for manual labelling.
Up to 50 files per batch, 2 GB per file and 5 GB per batch. A typical study round usually fits in one or two batches.
Plain text, Markdown, SubRip and WebVTT. Markdown is usually the easiest format to import into qualitative coding tools.
On the FLEURS English set (194 clips) the default model scored 5.93% word error rate in our benchmark. Conversational interviews with crosstalk, accents or background noise are harder than read speech — EDACC, the accented-speech set, scored 19.18%. Review quotes against the audio before publishing them.
No. It removes a third-party audio processor from the workflow, but you still own consent, retention and review requirements. Confirm the setup with your ethics board or data protection officer.
mp3, wav and m4a, English only in this beta. Video files cannot be transcribed directly.
Beta users get free transcription for good. Leave your email to reserve the beta — no payment, no card. We'll send the signed build to the same inbox once it's ready. No newsletter, no drip sequence, no sharing your address.
We sent a confirmation link. Click it to confirm your address — your reservation is complete, and the beta build goes to the same inbox the moment it's signed.