Security & Privacy
Exactly what happens to your audio and transcripts at every step. No marketing fluff — just how it actually works.
Three Levels of Privacy
Standard
Every user, every plan — including free
- HTTPS (TLS 1.3) for all data in transit
- Audio deleted immediately after processing
- Transcripts stored in our database
- We can read stored transcripts
- Data never sold or used for training
- Delete your data anytime
Private Transcript
Pro+ Available on Pro and Business plans
- Everything in Standard, plus:
- Transcript encrypted in your browser (AES-256-GCM) before saving
- We store only encrypted data — we cannot read it
- Key derived from your password, never sent to us
- ⚠ Audio is still processed on our servers during transcription
Private Cloud / Self-Hosted
Full isolation — from $99/mo
- Audio never leaves your infrastructure
- Transcription runs on your GPU
- No data sent to STT.ai servers
- Air-gapped support available
- True end-to-end privacy
What Actually Happens to Your Data
A transparent, step-by-step breakdown of how your audio and transcript are handled.
Standard (all users)
1
You upload audio or record live
Your file is sent over HTTPS (TLS 1.3) to our GPU server for transcription.
2
Audio is processed in memory
Our AI models transcribe your audio on the GPU. The audio is held in memory during processing — never written to disk — and deleted from memory immediately after.
3
Transcript is stored in our database
The text transcript, timestamps, and speaker labels are saved so you can access them later. We can read this data (this is how search, AI summaries, and sharing work).
4
You can delete everything anytime
Delete individual transcripts or your entire account from Privacy Settings. Deletion is permanent and immediate.
With Private Transcript enabled
Steps 1-2 are the same — your audio must be processed on our servers to generate the transcript. The difference is what happens next:
3
Transcript is encrypted in your browser before saving
After transcription, the result is returned to your browser. Your browser encrypts it with AES-256-GCM using a key derived from your password (PBKDF2, 100K iterations). The encrypted blob is then sent to our servers for storage. We never see or store the encryption key.
4
We store only encrypted data
Our database contains only the encrypted blob. We cannot decrypt it. If our database were breached, your transcripts would be unreadable.
Important: Private transcript protects the stored transcript. During the transcription process itself, your audio is processed on our servers in order to generate the text. If your threat model requires that audio never touches third-party servers, consider Private Cloud or Self-Hosted.
What We Can and Can't See
We CANNOT see (with Private Transcript)
- Your stored transcript text
- Speaker names or labels (stored)
- Timestamps or word-level data (stored)
- Your encryption key or password
We CAN see (even with Private Transcript)
- Your audio during processing (deleted after)
- File name, size, duration (metadata)
- Language detected, model used
- Timestamp of transcription
- Your account info and billing
Technical Details
| Encryption algorithm | AES-256-GCM (authenticated encryption) |
| Key derivation | PBKDF2 with SHA-256, 100,000 iterations |
| IV (nonce) | Random 12 bytes per encryption (never reused) |
| Key storage | Never stored — derived from password on each session |
| Transport encryption | TLS 1.3 (HTTPS) + HSTS (1 year, preload) |
| Audio retention | Processed in memory, never written to disk, deleted immediately |
| Implementation | Web Crypto API (browser-native, no external libraries) |
| Source code | github.com/sttaigit/stt-encryption (MIT license) |
Private Transcript Trade-offs
Private transcript is opt-in because encrypting the stored transcript limits some features:
Works with encryption
- Viewing your transcripts
- Exporting (TXT, SRT, VTT, etc.)
- Downloading
- Editing (decrypted in browser)
Not available with encryption
- Server-side search across transcripts
- AI summaries and chat (server can't read data)
- Public sharing via link
- Team workspace collaboration
Need Audio to Never Leave Your Servers?
Private transcript protects the transcript at rest, but audio still passes through our GPU during processing. If your compliance or security requirements demand that audio never touches third-party infrastructure, these are your options:
Private Cloud
$499/mo
Dedicated GPU server managed by us. Your audio never leaves your isolated environment.
- Dedicated A100 GPU
- Isolated — no shared infrastructure
- Audio processed on your hardware only
- Full API access + SLA
Self-Hosted
$99/mo
Docker image. Your servers. Your GPU. Nothing leaves your network.
- Docker — runs on any NVIDIA GPU
- Air-gapped support — no internet required
- Model updates included
- Full control, full privacy
Our Commitments (All Users, All Plans)
- Audio files are never stored permanently. Processed in GPU memory, deleted immediately after transcription.
- Your data is never used for AI training unless you explicitly opt in via Voice Lab.
- We don't sell your data. Ever. To anyone.
- All traffic encrypted in transit via TLS 1.3 with HSTS.
- Delete your data anytime from Privacy Settings or by deleting your account.
- Encryption code is open-source — audit it yourself (MIT license).
Open-Source Encryption
Our encryption library is fully open-source under the MIT license. Don't trust us — verify the code. No trust required, just math.
View on GitHub | View SourceReady to transcribe securely?
Upload your first file free. Private transcripts available on Pro and Business plans.
Start TranscribingUsein kysyttyjä kysymyksiä
Lataa ääni- tai videotiedosto STT.ai:een. Valitse haluamasi tekoälymalli ja -vaihtoehdot ja napsauta Transcribe. Selostus valmistuu muutamassa minuutissa. Vie TXT-, SRT-, VTT-, DOCX-, JSON- tai PDF-muodossa.
Kyllä! STT.ai tarjoaa 600 ilmaista minuuttia kuukaudessa kaikille käyttäjille. Ensimmäiseen transkriptioon ei tarvitse kirjautua. Maksetut suunnitelmat, joissa on enemmän minuutteja ja ominaisuuksia, alkavat 5 dollarista kuukaudessa.
Tarkkuus riippuu valitsemastasi tekoälymallista ja äänenlaadusta. Parhailla malleillamme saavutetaan 5–7 prosentin Word Error Rate -arvo viitearvoilla, mikä tarkoittaa 93–95 prosentin tarkkuutta. Selkeä ääni, jossa taustamelu on mahdollisimman vähäistä, tuottaa parhaat tulokset.
STT.ai tarjoaa 10+-malleja, kuten Whisper Large V3, NVIDIA Canary, ja enemmän. Voit verrata tuloksia eri malleista samassa tiedostossa.
Kyllä. Vie transkriptisi strt- tai VTT-tekstitystiedostoina, jotka toimivat YouTubessa, Vimeossa ja kaikilla suurilla videoalustoilla.
Kyllä. STT.ai tunnistaa ja leimaa automaattisesti eri kaiuttimet tekoälykaiuttimen diarisoinnin avulla. Toimii kaikilla malleilla ja kielillä.
Useimmat tiedostot on kirjoitettu alle 5 minuutissa. 1 tunnin äänitiedosto kestää yleensä 2-3 minuuttia nopeimmilla malleillamme.
STT.ai tukee 20+-ääni- ja videoformaatteja, kuten MP3, WAV, M4A, FLAC, OGG, MP4, MKV, MOV, WebM ja AVI. Vie TXT-, SRT-, VTT-, DOCX-, JSON- tai PDF-muodossa.
Kyllä. Äänitiedostoja käsitellään ja poistetaan transkription jälkeen. Tietojasi ei koskaan käytetä koulutukseen. Asiakaspuolen salaus on ilmainen kaikissa suunnitelmissa – se salaa tallennetut transkriptit avaimella, joka vain sinulla on. Käsittelyn aikana palvelin käsittelee ääntäsi tavallisella tekstillä. Tutustu turvallisuuteemme.
Kyllä. STT.ai tarjoaa REST-rajapintaa Pythonilla ja Node.js SDK:lla. Vapaa taso sisältää 100 minuuttia kuukaudessa.
Kyllä. STT.ai sisältää sisäänrakennetun transkriptioeditorin, jossa voit korjata virheitä, nimetä kaiuttimet uudelleen ja säätää aikaleimat.
Jokainen transkripti saa ainutlaatuisen jaettavan linkin. Vie DOCX:ään tai PDF:ään sähköpostia varten. Pron suunnitelmat tarjoavat salasanasuojattuja ja pysyviä linkkejä.
STT.ai tukee 1 300+ alustaa, kuten YouTubea, Vimeoa, TikTokia, SoundCloudia ja muuta. URL-transkriptio toimii vain julkisesti saatavilla olevalla äänen ja videon avulla. DRM-suojattua sisältöä (kuten Spotifyn premium-jaksoja, Netflixiä, Disney+:ta) ei voi siirtää. DRM-sisällön osalta lataa tiedosto erikseen ja lataa se suoraan.