ElBruno.QwenTTS 0.6.1

Qwen3-TTS ONNX Pipeline + C# .NET

Run Qwen3-TTS text-to-speech locally from C# using ONNX Runtime — no Python needed at inference time. Models are downloaded automatically on first run.

Pre-exported ONNX models are hosted on HuggingFace: elbruno/Qwen3-TTS-12Hz-0.6B-CustomVoice-ONNX (preset voices) | elbruno/Qwen3-TTS-12Hz-0.6B-Base-ONNX (voice cloning)

Features

• Local TTS Inference — Run Qwen3-TTS entirely on your machine using ONNX Runtime
• Automatic Model Download — Models download from HuggingFace on first run (~5.5 GB)
• Multi-Speaker — 9 built-in voices: ryan, serena, vivian, aiden, eric, dylan, uncle_fu, ono_anna, sohee
• Voice Cloning — Clone any voice from a 3-second audio sample (docs)
• Web UI — Blazor app with TTS generation and voice cloning pages (docs)
• GPU Acceleration — Optional CUDA or DirectML support via SessionOptions injection (docs)
• Multi-Language — English, Spanish, Chinese, Japanese, Korean
• Voice Control — Instruction-based style (e.g., "speak with excitement")
• Shared Model Cache — Models stored once in %LOCALAPPDATA%/ElBruno/QwenTTS, shared across all apps
• 24 kHz WAV Output — High-quality mono audio

Quick Start

Install via NuGet

dotnet add package ElBruno.QwenTTS

Generate speech in C#

using ElBruno.QwenTTS.Pipeline;

// Models download automatically on first run (~5.5 GB)
using var pipeline = await TtsPipeline.CreateAsync("models");
await pipeline.SynthesizeAsync("Hello world!", "ryan", "hello.wav", "english");

CLI

dotnet run --project src/ElBruno.QwenTTS -- --model-dir models --text "Hello, this is a test." --speaker ryan --language english --output hello.wav

Models are downloaded automatically if not present in the --model-dir directory.

Voice Cloning

Clone any voice from a 3-second audio sample using the ElBruno.QwenTTS.VoiceCloning package:

dotnet add package ElBruno.QwenTTS.VoiceCloning

using ElBruno.QwenTTS.VoiceCloning.Pipeline;

var cloner = await VoiceClonePipeline.CreateAsync();
await cloner.SynthesizeAsync("Hello world!", "reference_speaker.wav", "output.wav", "english");

See docs/voice-cloning.md for full documentation.

GPU Acceleration

Pass a sessionOptionsFactory to use CUDA or DirectML instead of CPU:

using ElBruno.QwenTTS.Pipeline;

// CUDA (NVIDIA) — requires Microsoft.ML.OnnxRuntime.Gpu NuGet package
var tts = await TtsPipeline.CreateAsync(
    sessionOptionsFactory: OrtSessionHelper.CreateCudaOptions);

// DirectML (any GPU on Windows) — requires Microsoft.ML.OnnxRuntime.DirectML NuGet package
// Uses GPU for language model, CPU for vocoder (hybrid mode)
var tts = await TtsPipeline.CreateAsync(
    sessionOptionsFactory: OrtSessionHelper.CreateDirectMlOptions,
    vocoderSessionOptionsFactory: OrtSessionHelper.CreateCpuOptions);

See docs/gpu-acceleration.md for full setup instructions.

More Examples

dotnet run --project src/ElBruno.QwenTTS -- --model-dir models --text "Welcome to the future of speech synthesis." --speaker serena --output welcome.wav
dotnet run --project src/ElBruno.QwenTTS -- --model-dir models --text "Speaking with excitement and energy!" --speaker aiden --instruct "speak with excitement" --output excited.wav
dotnet run --project src/ElBruno.QwenTTS -- --model-dir models --text "A calm and gentle narration." --speaker ryan --instruct "speak slowly and calmly" --output calm.wav

Spanish Examples

dotnet run --project src/ElBruno.QwenTTS -- --model-dir models --text "Hola, esta es una prueba de texto a voz." --speaker ryan --language spanish --output hola.wav
dotnet run --project src/ElBruno.QwenTTS -- --model-dir models --text "Bienvenidos al futuro de la sintesis de voz." --speaker serena --language spanish --output bienvenidos.wav

File Reader (batch audio from text/SRT files)

dotnet run --project src/ElBruno.QwenTTS.FileReader -- --model-dir models --input samples/hello_demo.txt --speaker ryan --language english --output-dir output/hello
dotnet run --project src/ElBruno.QwenTTS.FileReader -- --model-dir models --input samples/demo_subtitles.srt --speaker serena --output-dir output/subtitles

Web App (browser UI)

dotnet run --project src/ElBruno.QwenTTS.Web

Open http://localhost:5153 — two pages:

• 🔊 TTS — type text or upload files, pick a voice, and generate speech
• 🎭 Voice Clone — record your voice or upload a WAV, then synthesize with your cloned voice

Documentation

Python Tools

The python/ directory contains tools for exporting ONNX models from PyTorch weights and downloading models from HuggingFace. These are only needed if you want to re-export or customize models — they are not required for running the C# pipeline.

Building from Source

git clone https://github.com/elbruno/ElBruno.QwenTTS.git
cd ElBruno.QwenTTS
dotnet build
dotnet test

Requirements

• .NET 8.0 or .NET 10.0 SDK
• ONNX Runtime compatible platform (Windows, Linux, macOS)
• ~5.5 GB disk space for model files

Contributing

Contributions are welcome! Here's how to get started:

1. Fork the repository
2. Create a branch for your feature or fix: git checkout -b feature/my-feature
3. Make your changes and ensure the solution builds: dotnet build
4. Run tests: dotnet test
5. Submit a pull request with a clear description of the changes

Please open an issue first for major changes or new features to discuss the approach.

References

• Qwen3-TTS GitHub
• Original model (PyTorch)
• Pre-exported ONNX models

👋 About the Author

Hi! I'm ElBruno 🧡, a passionate developer and content creator exploring AI, .NET, and modern development practices.

Made with ❤️ by ElBruno

If you like this project, consider following my work across platforms:

• 📻 Podcast: No Tienen Nombre — Spanish-language episodes on AI, development, and tech culture
• 💻 Blog: ElBruno.com — Deep dives on embeddings, RAG, .NET, and local AI
• 📺 YouTube: youtube.com/elbruno — Demos, tutorials, and live coding
• 🔗 LinkedIn: @elbruno — Professional updates and insights
• 𝕏 Twitter: @elbruno — Quick tips, releases, and tech news

License

This project is licensed under the MIT License — see the LICENSE file for details.