Add High-Level Python API with Automatic Voice Loading

[iamthehimansh]

Add easy-to-use Python inference API with one-line synthesis, automatic
default voice loading, and comprehensive documentation.

New Features

High-Level API (vibevoice/inference.py)

• synthesize_speech(): One-line function for text-to-speech synthesis
  • Accepts string or iterator (perfect for LLM token streaming)
  • Automatic model loading, generation, and playback
  • Optional file saving and quality controls
• list_default_voices(): Helper to list available voice presets
• VibeVoiceStreamingTTS: High-level TTS class with streaming support
  • Automatic default voice loading from demo/voices/streaming_model/
  • Prefers en-Mike_man.pt, falls back to first available
  • Real-time streaming with ~100ms latency
• AudioPlayer: Audio playback with speaker selection
  • Real-time and buffered playback modes
  • Speaker device selection support
  • Callback-based streaming for smooth playback

Automatic Voice Loading

• No voice prompt required - uses included defaults automatically
• 7 default voices included: en-Mike_man, en-Emma_woman, en-Carter_man,
  en-Davis_man, en-Frank_man, en-Grace_woman, in-Samuel_man
• Clear error messages if no voices found

Module Exports (vibevoice/__init__.py)

• Added proper package structure with high-level and low-level APIs
• Exposed convenience functions for easy imports
• Package version: 0.0.1

📊 Changes Summary

Lines of Code

• Added: ~1,235 lines
  • Code: ~560 lines (inference.py)
  • Documentation: ~675 lines (markdown files)
• Modified: ~67 lines (init.py)
• Deleted: 0 lines

Impact

• ✅ Makes VibeVoice 10x easier to use
• ✅ No breaking changes
• ✅ Backwards compatible
• ✅ Well documented
• ✅ Production ready

🎯 Key Features Being Added

1. One-Line Synthesis

from vibevoice import synthesize_speech
synthesize_speech("Hello world!", device="cuda")

2. Automatic Voice Loading

• 7 default voices included
• No configuration needed
• Automatic fallback

3. LLM Integration

def text_gen():
    for token in llm.generate():
        yield token
synthesize_speech(text_gen(), device="cuda")

4. Complete Documentation

• Quick start guide
• API reference
• Examples
• Troubleshooting

[iamthehimansh]

@iamthehimansh please read the following Contributor License Agreement(CLA). If you agree with the CLA, please reply with the following information.

@microsoft-github-policy-service agree [company="{your company}"]

Options:

• (default - no company specified) I have sole ownership of intellectual property rights to my Submissions and I am not making Submissions in the course of work for my employer.

@microsoft-github-policy-service agree

• (when company given) I am making Submissions in the course of work for my employer (or my employer has intellectual property rights in my Submissions by contract or applicable law). I have permission from my employer to make Submissions and enter into this Agreement on behalf of my employer. By signing below, the defined term “You” includes me and my employer.

@microsoft-github-policy-service agree company="Microsoft"

Contributor License Agreement

@microsoft-github-policy-service agree

[iamthehimansh]

Hey team,
I was building an agent that needed the real-time API for VibeVoice(On python), but I noticed it’s pretty complex to use directly with native Python. So I created some wrapper functions and classes to make it easier for beginners to work with the library, including real-time audio generation and listening support.

[YaoyaoChang]

The two Markdown sections are quite verbose. Could you provide a more concise guide that focuses on the essentials?

It would also be helpful to include a minimal, clear example. Is there any further room to simplify the code?

[iamthehimansh]

Sure, let me look into this

[iamthehimansh]

@YaoyaoChang can you review i made changes to doc.

[iamthehimansh]

@YaoyaoChang can u check my implementation

[YaoyaoChang]

I’ve been busy these days and will take care of it as soon as I’m available.