Skip to content
Sign in

Give yourapplication a voice.

Turn text into speech and speech into text. Run locally. Embed it in your application.

0:00

Your speech workspace.

Choose a voice, edit your text and listen to the result.

SSML editor, word tracking and audio waveform in the SintelSES demo.

Text to speech (TTS)

Write it. Hear it.

Read application messages, dialogue prompts and SSML with local speech models. Supported engines can deliver audio before the full utterance is complete.

  • Decoder, text-chunk or autoregressive streaming, depending on the model
  • Languages and voices follow the installed model packages
  • SSML preparation, reading rules, pauses and speaking speed
Text to speech preview
0:00

SSML support

Shape how your words are spoken.

Use SSML to control pauses, speaking speed and how numbers are read. Supported tags and languages depend on the selected model.

  • Add timed pauses between sentences with break
  • Change the speed of a passage with prosody
  • Choose how numbers, dates and letters are read with say-as
Prerecorded SSML sample · Sonik
0:00
From audio to readable text

Give your application a voice with Sintel SES

Turn text into speech, and speech into text

Run speech models on your own server, with your own applications and telephony workflows.

Your audio, your infrastructure.Try transcription

Speech to text (STT)

Speak. See it.

Transcribe recordings and live audio on your server. Mira and Nemo support native streaming; Duyum supports file transcription and revisable text previews.

  • Transcribe WAV recordings and recorded microphone audio
  • Native live decoding with streaming-capable engines
  • Mix, left, right or separate channel processing in the speech tools

Voice gallery

Choose the voice that fits.

A selection from your installed voices. Sign in to explore the complete library and read your own text.

İdil
Ahen
tr
Sign in to listen

Only installed models and their reported languages are shown. Available voices and capabilities depend on the selected model and package.

Architecture \u0026 deployment

Run speech on your infrastructure.

Embed speech models in your application or connect through an API, telephony integration or local service. Generate speech and transcribe audio on the CPU, with no GPU required.

Embedded in your application
Load models inside your application with the SDK and run speech generation and recognition in the same process. Manage requests and audio streams from your application.
SDK · In-process execution
MRCPv2
Provide speech synthesis and recognition resources to telephony and IVR systems through MRCPv2. Examples of MRCPv2-capable platforms include FreeSWITCH and Asterisk through UniMRCP, Cisco CVP and Avaya Experience Portal.
SIP · MRCPv2 · RTP
Unix socket
Run the speech service as a local daemon. Send requests over a Unix socket and receive paced PCM audio.
PCM · local IPC
OpenAI API
Use the documented TTS and ASR endpoints with OpenAI clients. See the API guide for supported audio formats.
HTTP · TTS · STT
SSML support
Control pauses, speaking speed and explicit reading rules with SSML. Supported tags and languages depend on the model.
break · prosody · say-as
Runs on the CPU
Speech generation and recognition require no GPU. Installed models run locally on the CPU; performance depends on the model and processor.
No GPU required

Speech inference runs on your server. Install model packages once, then process audio locally. Downloading models may require an internet connection.

Interactive studio

Experience SintelSES in your browser.

A voice library, text composer and transcription workspace await after sign-in. The preview below introduces the available tools.

Model-aware controls

  • Choose a supported reading language
  • Select a voice from the installed catalog
  • Adjust speed when the model supports it
  • Listen, inspect the waveform and download WAV
0:00

Installed engines

Model and capability comparison.

Only installed models and their reported languages are shown. Available voices and capabilities depend on the selected model and package.

ModelTaskLanguagesSample rateCapabilities
AhenTTS
tr
48 kHzLocal speech synthesis using an installed model.
Mira

Powered by mihu

ASR
tr
16 kHzNative streaming transcription for live audio.

Measured locally · docs/BENCHMARKS.md

Performance, with context.

A recorded Duru 3.1 (Vocos-ft) streaming measurement. These are development-machine results, not a latency guarantee.

First PCM

10.96 msDuru · Stream

Generation time

45.02 msDuru · Stream

Real-time factor

0.01119Duru · Stream

Audio duration

4.022 sDuru · Stream

Apple M4 Max · macOS arm64 · ONNX Runtime 1.30.0 CPU · 4 threads · 1 warm-up + 3-iteration average · October 17, 2026

Studio voice, Turkish, 24 kHz; warm Stream with a non-blocking callback, speed=1, seed=42. Model opening, network, WAV writing and playback are excluded. First PCM is not audible latency or network RTT.

Built for developers

From your first import to your first call.

Use the Go SDK in your application or connect to the local service. The examples use the current SDK and CLI interfaces.

1

Install your models

Prepare the runtime and the language/model packages for your target machine.

2

Start the local engine

Open a model in Go or run the SintelSES service on your host.

3

Connect your application

Choose the SDK, Unix socket or MRCPv2 integration. The SDK currently uses a local module replace.

OpenAI API

Use the documented TTS and ASR endpoints with OpenAI clients. See the API guide for supported audio formats.

Sign in to read the API guide