How transcription APIs handle your audio
Loquela Remote keeps no audio and no transcripts, never trains on them, and turns speech into text on hardware we own. Here is what other speech-to-text APIs do by default, read from each provider's own pages on 13 November 2026.
Provider by provider
| Provider | Keeps your audio or transcript by default | Can use it to improve its models by default | A way to keep nothing | Where your data is kept | Source |
|---|---|---|---|---|---|
| Loquela Remote | No, never | No, never | Built in, for every customer | Nothing is kept. Transcribed on our own hardware in North Carolina | Your audio |
| OpenAI | No for transcription, per its endpoint table | No, unless you opt in | Zero Data Retention, with OpenAI's approval | Regional options in its data residency guide | openai.com |
| AssemblyAI | Audio up to 48 hours, transcripts 30 days | Yes, unless you opt out | For streaming and sync, after opting out of training | US, or its European servers | assemblyai.com |
| Deepgram | Some, unless you opt out on each request | Yes, unless you opt out on each request | Opt out on each request | Not stated | deepgram.com |
| ElevenLabs | Yes by default (zero retention is for Enterprise) | Not stated | Zero Retention Mode, for select enterprise customers | Not stated | elevenlabs.io |
| Groq | Can log audio and transcripts for up to 30 days | Not stated | Zero Data Retention, once an admin turns it on (batch jobs then stop working) | On Google Cloud in the United States | groq.com |
Defaults for standard API accounts. "Not stated" means the provider's page does not say.
Questions worth asking any provider
- Is the audio kept, and for how long? Some providers delete audio within two days and keep transcripts for a month.
- Are transcripts kept separately? A transcript can hold as much as the recording.
- Is training on by default? And is opting out a setting on your account, or a flag you must send with every request?
- What do logs keep? Abuse-monitoring and debugging logs can hold content after the request ends.
- Who else touches it? Ask where the audio is processed and which other companies process it.
- Can it be encrypted end to end? So the networks in between carry only ciphertext.
How Loquela Remote answers them
- Nothing kept. Audio and transcripts stay in memory and are erased when your request ends. We keep one line per request: a request number, where it came from, how long the audio was, the minutes charged, whether it worked and how long it took.
- Nothing to train on. With nothing kept, there is nothing to train on, by default or otherwise.
- No other AI company. Transcription runs on hardware we own in North Carolina.
- PGP on Pro and Business. Encrypt a recording to your account's key before it leaves your computer and get the transcript back encrypted to yours. See Security.
Each step a recording takes is on How your audio is handled. Prices are on Pricing.