Endpoints | Prerecorded
Transcription API that
gets the words right
Upload a complete audio file. Get the full transcript back in one response, with the medication names, order numbers and case references that generic models mishear.
Every file transcribed in your domain language
Built for batch work where accuracy matters more than speed: consultations, calls and recordings that need a complete, correct record.
Your vocabulary, from day one
Upload your terms as text and Reson8 writes product names, identifiers and jargon the way your team does. No audio samples, no fine-tuning.
Hour-long recordings, one request
Send the file, get JSON back. Meetings, calls and archives that run past an hour come back attributed by speaker.
EU native, zero retention
Audio runs on Reson8's own GPUs in the Netherlands and is never stored or used for training. ISO 27001 certified.
Send a file.
Get the transcript.
The response carries the transcript, the detected language and speaker labels. Add your own vocabulary and the engine adapts to it.
10 European languages, one endpoint
Start from your AI assistant
Teams that ship voice features on Reson8
Financial advice
Obsidian
Adviser-client meetings transcribed inside Obsidian's own AWS environment. About 4 weeks from first test access to signed agreement.
AI platform
SafeGPT
Recordings for 1,250+ Dutch organisations, on GPUs in the Netherlands. 4 weeks from first meeting to tested integration.
Video
VEED
Word-level timestamps on user-uploaded video power subtitles, animated text and filler-word removal. About 5 weeks to live.
Frequently asked questions
Reson8 is built for European languages and gets the words right that matter most. Generic speech models mishear medication names, order numbers, addresses and case references. Reson8’s speech-to-text adapts to your domain-specific terminology in real time or from a single text upload, no fine-tuning or audio samples needed.
We have full support for 10 languages: English, German, Spanish, French, Dutch, Italian, Polish, Portuguese, Swedish, Frisian.
Reson8 routinely leads speech-to-text benchmarks from third-parties such as Hugging Face, Artificial Analysis, Coval, and Voice Arena.
Yes, a single stream can switch dynamically between languages mid-conversation. Language detection is automatic and out-of-the-box, no manual language flagging necessary. Find more information on language detection here.
Yes, diarization is supported across all endpoints, ensuring transcripts accurately distinguish between speakers no matter the complexity of the audio. Find more information on diarization here.
Each endpoint is applicable for a different use case:
- Prerecorded: for complete audio files, uploaded over REST. You get the full transcript back in a single response.
- Realtime: for live audio streamed over WebSocket, when you want to show live transcripts while speaking.
- Turns: also a WebSocket stream, but it detects when a speaker starts and stops talking, returning the transcript of each turn separately. Use this endpoint for cases like voice agents, when you need low latency end-of-turn detection and native barge-in.
Check out our guide on choosing the right endpoint here.
No, we don't and never will. Audio is processed on our own GPU infrastructure, which is fully built and operated by Reson8. Find more information on our trust center here.
Reson8 retains zero data. All audio is streamed pass-through and never stored; this also applies to files and transcripts.
Yes, Reson8 is fully ISO 27001 certified and GDPR compliant. We own and operate our full GPU infrastructure ourselves in the Netherlands, Reson8 is not reliant on any third party subprocessors of audio, making Reson8 uniquely suitable to regulated industries like healthcare, legal, finance, and more. Find our DPA here.
Sign up today! Get an API key through console.reson8.dev, talk to our team or check our documentation to get started!
Uploaded file formats are detected automatically. For raw audio, set the encoding parameter. See Audio Formats in the docs.
Yes. Reson8 deploys into your VPC with a Helm chart, which is how Obsidian runs it on AWS.