Soft Leaf
Using ReadSpeaker's TTS plugin saved us countless hours in development and brought us closer to our vision of a truly accessible game.
Read the full Soft Leaf storyReadSpeaker speechEngine SDK
On-device, offline text-to-speech SDK for desktop and PC apps
Install the speech engine right on the target machine, where it generates audio locally with no network calls. 150+ AI voices across 50+ languages, with native bindings for C, C++ and Java on Windows and Linux, and C# on Windows.
- C/C++ · C# · Java
- Windows 10/11 + Linux
- On-device
Customer story
Built into shipping software, on device
Features
On-device voice, built for developers
Everything your team needs to put natural-sounding voice inside a desktop or PC application, with the engine running locally.
Ship voice that never leaves the device
The engine runs on your customer's machine and generates audio with zero network calls. A good fit for air-gapped, regulated or offline-first deployments where the data stays in-house.
150+ AI voices in 50+ languages
Natural, human-sounding AI voices, with new voices and languages added to the catalog regularly. Pick the accent and tone that fit each market, and ship them with a footprint that stays light on disk.
Your team writes in their stack
Native bindings for C and C++ on Windows and Linux, C# on Windows, and Java on Windows and Linux. Sample code, documentation and implementation support from our team make it easy to integrate.
Fine-grained voice control with SSML
Switch voice and language mid-sentence, control prosody, insert pauses and phonetic transcriptions. Speech Synthesis Markup Language (SSML) input is fully supported, in UTF-8.
Speak brand names and jargon correctly
Build per-language pronunciation dictionaries for product names, acronyms and proper nouns, with multiple dictionaries per language. Use IPA transcription when you need precise phonetic control.
Audio formats for any pipeline
16-bit linear PCM, 8-bit A-law and μ-law for telephony, Dialogic ADPCM for legacy IVR. Output the format your pipeline expects, raw or wrapped in WAV.
Tune every utterance
Dial speaking rate, pitch and volume to match the moment, from calm narration to urgent alerts. Configure them as needed for each deployment.
Predictable licensing
A license file installed alongside the SDK locks the voices and term you ordered. No surprise upgrades, no usage-based billing, no audit anxiety.
Real humans help you ship
ReadSpeaker's support team works with your developers through integration. Sample code, documentation and direct contact mean you're never stuck reading a wiki at 2 AM.
Try it in your stack
A simple C++ integration example
Native bindings ship for C and C++ on Windows and Linux, C# on Windows, and Java on Windows and Linux. Call the engine in-process and synthesise text to a local audio file, with no network calls.
// Representative example — the exact API and headers ship with the licensed SDK.
#include "speechengine.h"
int main() {
SpeechEngine engine;
engine.init("./license.lic"); // license file enforces voices + term
engine.selectVoice("en_US-neural"); // 150+ voices, 50+ languages
engine.setSpeechRate(90); // tune rate, pitch and volume per request
// Synthesise plain text or SSML to a local audio file — no server calls
engine.synthesizeToFile("Welcome to ReadSpeaker.", "welcome.wav");
engine.release(); // free the engine
return 0;
}Technical specifications
Every spec your dev and ops teams will ask about
Sized for on-device, offline deployment: small footprint, broad OS support, telephony-ready audio.
| Specification | Detail |
|---|---|
| Programming languages | C / C++ (Windows/Linux) · C# (Windows) · Java (Windows/Linux) |
| Windows OS | Windows 10 and 11 |
| Linux OS | CentOS Stream 9.x to 10.x · Ubuntu 18.x to 24.x · RHEL 7.x to 9.x · Other OS on request |
| CPU | Intel x86/64, 1 GHz or higher. AVX required on x64 architectures (contact ReadSpeaker if your environment lacks AVX support). |
| RAM | 6 to 30 MB |
| Voice footprint | 9 to 35 MB per voice (Deep Neural Networks technology) |
| Runtime memory | 20 to 30 MB (Neural Standard) |
| Input formats | Plain text (multibyte) · SSML (UTF-8) |
| Audio output formats | 16-bit linear PCM (raw/WAV) · 8-bit PCM A-law/μ-law (raw/WAV) · 8-bit unsigned linear PCM Wave · 4-bit Dialogic ADPCM |
| Documentation | Product Specification (PDF) |
Deployment model
Runs on your infrastructure, under your license file
speechEngine SDK is installed locally on the target machine. The TTS engine runs without server calls. License agreement and license file enforce the licensed platform, voices and term. For corporate compliance documentation (ISO/IEC 27001, GDPR, EU SCCs), see the ReadSpeaker Privacy Notice and About pages.
On-device (Engine installed on the target machine)
Installed locally
No network calls (Audio generated without server round-trips)
Engine runs locally
License file enforcement (License file delivered with the SDK)
Voices + term
SDK delivery (SDK built for the platform on the order)
Per customer order
FAQ
Questions about speechEngine SDK
speechEngine SDK is an on-device text-to-speech SDK for developers building desktop and PC-based applications. The engine installs locally on the target machine and supports 150+ AI voices across 50+ languages. It runs without network connectivity, which suits air-gapped and privacy-sensitive environments.
speechEngine SDK ships native bindings for C and C++ on Windows and Linux, C# on Windows, and Java on Windows and Linux. The SDK is delivered for the platform specified in the customer's order.
speechEngine SDK runs on Windows 10 and 11, and on Linux distributions including CentOS Stream 9.x to 10.x, Ubuntu 18.x to 24.x, and RHEL 7.x to 9.x. Other operating systems are supported on request. CPU: Intel x86/64 1 GHz or higher (AVX required on x64). RAM: 6-30 MB.
Yes. The TTS engine installs locally on the target machine and generates audio with no network calls, which suits privacy-sensitive and air-gapped deployments. For ARM, mobile or IoT targets, see speechEngine SDK Embedded.
speechEngine SDK outputs four audio formats: 16-bit linear PCM (raw/WAV), 8-bit PCM A-law/μ-law (raw/WAV), 8-bit unsigned linear PCM Wave, and 4-bit Dialogic ADPCM. A-law and μ-law output is particularly suited for telephony integration.
Yes. speechEngine SDK accepts plain text input (multibyte) and SSML input (UTF-8), so your application can mark up the speech it needs. SSML lets you switch voice and language within a single text, modify prosody, insert pauses, and apply phonetic transcriptions.
speechEngine SDK is governed by a license agreement that defines the licensed platform, voices, and permitted use context. A license file delivered with the SDK technically enforces the licensed voices and the term, meaning the duration of permitted use.
Yes. Implementation support from the ReadSpeaker Support Team is included. The team assists customers in understanding and carrying out the integration according to the provided documentation and sample code.