Reson8 Turn-Based STT Now Available for Telnyx Voice AI

Reson8 is now available as a speech-to-text engine on Telnyx, adding turn-based transcription of ten European languages, with automatic language detection, to Telnyx Voice AI. Select reson8/turns in Mission Control or through the API, and keep the same Call Control, WebSocket, TeXML, and AI Assistants configuration every other Telnyx STT engine uses.

What's new

• Turn-based STT engine: reson8/turns delivers transcripts per turn of speech. Interim frames carry the full transcript of the turn so far and supersede each other, and the final transcript arrives when the turn ends.
• Ten languages with automatic detection: Dutch, English, French, Frisian, German, Italian, Polish, Portuguese, Spanish, and Swedish. Set language to auto to detect the language, or fix it to nl, en, fr, fy, de, it, pl, pt, es, or sv.
• Available on WebSocket, Voice API, TeXML, and AI Assistants: Set the engine to Reson8 and the model to reson8/turns on WebSocket streaming, Call Control transcription_start, TeXML <Gather> and <Transcription>, or the AI Assistants transcription settings.
• Turn thresholds in Mission Control: The assistant's transcription settings show Eager Turn Probability (default 0.5) and Final Turn Probability (default 0.92), so you can see where a turn produces an interim candidate and where it closes.
• Raw audio input: Accepts linear16, linear32, mulaw, and alaw at any sample rate. Container formats are not supported. Transcript frames carry no confidence score, and endpointing is ignored.

Why it matters

• Another engine on the same integration surface. Teams that already run Telnyx STT can try Reson8 by changing the engine and model, with no new SDK or webhook.
• Voice agents that wait for a complete caller turn get one cumulative transcript per frame, so the agent does not stitch together word fragments to decide when to answer.
• Mixed-language call traffic across Dutch, French, German, Italian, Polish, Portuguese, Spanish, Swedish, and English can run through one engine with auto, with no per-call language label.
• Frisian is on the supported list, for assistants that serve Friesland.

Example use cases

• Voice agents for support lines that take calls in Dutch, French, and English.
• Contact centers in mixed-language European markets that want one STT engine and automatic language detection per call.
• Appointment, order, and delivery assistants that answer after the caller finishes a turn.
• Teams comparing STT engines on their own call audio by swapping the model on an existing assistant.

Getting started

1. In Mission Control, go to AI > Assistants and open the assistant you want to configure.
2. Select the Voice tab and scroll to Transcription.
3. Set the Transcription Model to reson8/turns and leave Transcription Language on Auto, or pick one of the ten languages.
4. Review Eager Turn Probability and Final Turn Probability, then save the assistant and test with a call.

To set it on an assistant through the API, use:

"transcription": {
  "model": "reson8/turns",
  "language": "auto"
}

For Call Control, POST to transcription_start:

{
  "transcription_engine": "Reson8",
  "transcription_engine_config": {
    "transcription_engine": "Reson8",
    "transcription_model": "reson8/turns",
    "language": "auto"
  }
}

For WebSocket streaming:

wss://api.telnyx.com/v2/speech-to-text/transcription?transcription_engine=Reson8&model=reson8/turns&input_format=linear16&sample_rate=16000

Each interim frame carries the full transcript of the turn so far, with no word-by-word partials, and the final frame arrives when the turn ends.

Learn more in the Speech-to-Text models docs, the AI Assistants transcription settings, or the WebSocket streaming parameters.