Text-to-Speech Preview
Speechmatics developed its own text-to-speech models for interactive speech. Its October 2025 development account describes a preview built around low-latency streaming.
Preview
Own model development
The company identifies its own models and infrastructure as the basis of this service. It connects the work to its experience with acoustic modeling, pronunciation and prosody. It does not publish an architecture, parameter count or separately downloadable checkpoint in the reviewed material.Speechmatics: Why we built our text-to-speech models
The current quickstart supports streaming output, but not bidirectional streaming. It also says speed, pitch and emphasis controls are unavailable. The four named voices are selectable outputs, not four documented foundation-model releases.Speechmatics documentation: Text-to-speech preview
Practical guidance. Audio can start playing before the response finishes. A client still needs to decide how to send text and handle an interrupted reply; streamed output alone does not solve conversational turn handling.Speechmatics documentation: Text-to-speech preview
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Access | API, Provider product |
| Model identifiers | Speechmatics TTS |
| Voices | Sarah and Theo: UK English; Megan and Jack: US EnglishSpeechmatics documentation: Text-to-speech preview |
| Output | 16 kHz, 16-bit mono audio; WAV or little-endian PCMSpeechmatics documentation: Text-to-speech preview |
| Access | Hosted preview API; own-environment deployment by agreementSpeechmatics documentation: Text-to-speech preview |
Related models
- Linden 1
Speechmatics · Linden
Speech recognition model for short voice-agent turns and spoken identifiers.
Launch date API - On-Device Speech Recognition
Speechmatics · On-Device
Local speech recognition model family for software running on user devices.
Launch date not confirmedRestricted accessProvider product