Voices Enables Fast Text-to-Speech in Java Applications

Voices: Fast Text-to-Speech for Java Applications

Date: 2025-11-16 10:16 Zhejiang

Voices enables Java applications to quickly implement text-to-speech (TTS) conversion.

---

Overview

Voices is an open-source TTS project designed for Java 17+.

Key features:

• No external APIs required.
• No manual installation of additional software.
• Supports multiple languages via dictionaries or OpenVoice.
• Generates audio files directly within Java applications.

Background

• Creator: Henry Coles — also known for Pitest and Arcmutate mutation testing.
• First introduced on Bluesky: September 2025.
• Latest release: Version 0.0.8 (October 2025).

Technology

• Built on ONNX Runtime — a cross-platform AI engine optimized for both training and inference.
• Supports models from TensorFlow, PyTorch, and more.
• Can accelerate on hardware processors where available.

---

Getting Started

Maven Configuration

Include the following dependencies in your `pom.xml`:

    org.pitest.voices
    chorus
    0.0.8




    org.pitest.voices
    alba
    0.0.8




    org.pitest.voices
    en_uk
    0.0.8




    com.microsoft.onnxruntime
    onnxruntime
    1.22.0

Tips:

• Replace `en_uk` with `en_us` for American English pronunciation.
• Swap `onnxruntime` for `onnxruntime_gpu` to enable GPU acceleration.
• Use a single `Chorus` instance to minimize model loading overhead.

---

Example: English Text to Audio

ChorusConfig config = chorusConfig(EnUkDictionary.en_uk());
try (Chorus chorus = new Chorus(config)) {
    Voice alba = chorus.voice(Alba.albaMedium());
    Audio audio = alba.say("This is the InfoQ article about the Voices library");
    Path path = Paths.get("InfoQ_English");
    audio.save(path);
}

---

Runtime Model Download

Add the downloader dependency:

    org.pitest.voices
    model-downloader
    0.0.8

Available factory classes:

• `org.pitest.voices.download.Models`
• `org.pitest.voices.download.UsModels`
• `org.pitest.voices.download.NonEnglishModels`

Example: Dutch Text to Speech

Model nlModel = NonEnglishModels.nlNLRonnie();
ChorusConfig config = chorusConfig(EnUkDictionary.en_uk());
try (Chorus chorus = new Chorus(config)) {
    Voice alba = chorus.voice(nlModel);
    Audio audio = alba.say("Dit is een Nederlandse tekst Scheveningen");
    Path path = Paths.get("Dutch");
    audio.save(path);
}

---

Using OpenVoice (No Dictionaries Required)

Dependency

    org.pitest.voices
    openvoice-phonemizer
    0.0.8

Note:

• File size: 50MB (vs. 3MB for dictionary).
• Higher computational requirements.

Example

ChorusConfig config = chorusConfig(Dictionaries.empty())
        .withModel(new OpenVoiceSupplier());
try (Chorus chorus = new Chorus(config)) {
    Voice alba = chorus.voice(Alba.albaMedium());
    Audio audio = alba.say("This is the InfoQ article about the Voices library");
    Path path = Paths.get("InfoQ_English_OpenVoice");
    audio.save(path);
}

---

GPU Support

To run on GPU:

• Remove `onnxruntime` dependency.
• Add `onnxruntime_gpu` dependency.
• Use `gpuChorusConfig`:

ChorusConfig config = gpuChorusConfig(EnUkDictionary.en_uk());

Default: GPU 0.

Custom CUDA options: use `.withCudaOptions()` in `ChorusConfig`.

---

Pause & Markdown Handling

• Markdown characters (`#`, `---`, em or en dashes) trigger pauses in audio output.
• Pause behavior can be customized via `ChorusConfig`.

---

Alternatives

Other TTS libraries like Sherpa Onnx or MaryTTS may have:

• More difficult Maven integration.
• Lower audio quality.

---

Interview Highlights: Henry Coles on Voices

Common Use Cases

> “Whenever you need to quickly generate natural-sounding speech without external services.”

Motivation

> “I needed a Java-native way to run piper models without relying on Python services.”

Challenges

• No linguistics background.
• Porting TypeScript logic to Java.
• Handling phonemisation exceptions with dictionaries.

Future Improvements

• Cleaner API design.
• Adjustable pause handling and speech pace.

Testing Recommendation

• Focus on input boundary tests.
• Moderate output tests to confirm audio generation.

---

Related AI Content Monetization

Platforms like AiToEarn官网 integrate:

• AI content generation.
• Cross-platform publishing (Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, X).
• Analytics & AI Model Rankings.

This ecosystem complements Voices by enabling multi-channel distribution of AI-generated speech content.

---

References

• Original Article: Voices Enables Fast Text-to-Speech for Java Applications
• Disclaimer: This is an InfoQ translation — reproduction without permission is prohibited.

---

Recommended Reads

• Cursor's $200B Valuation & Founder's Secrets
• Baidu Unveils Self-Evolving Intelligent Agent
• Meta's Internal Competition & Yann LeCun's Exit
• Jeff Barr on AI's Impact on Developers

---

Would you like me to also prepare a concise Quick Start guide version of this Voices documentation, so developers can implement TTS in under 5 minutes? That could make this Markdown even more actionable.