Unity Vosk Speech-to-Text for Apple Silicon

This project implements real-time speech-to-text functionality in Unity using the Vosk speech recognition toolkit, optimized for Apple Silicon Macs.

Features

• Real-time speech recognition
• Support for multiple microphones
• Customizable speech detection parameters
• Threaded audio processing for improved performance

Prerequisites

• Unity 2022.3 or later
• macOS with Apple Silicon (M1 chip or later)
• Xcode (for building on macOS)

Setup

1. Clone this repository or download the project files.

2. Open the project in Unity.

3. Download the Vosk model:

   • Go to the Vosk Models page
   • Download the vosk-model-small-en-us-0.15 model (or another model of your choice)
   • Extract the downloaded model to Assets/StreamingAssets/models/

4. Ensure the Vosk library is properly set up:

   • Check that libvosk.dylib is present in Assets/Plugins/macOS/
   • Verify that the VoskLoader.cs script is in your project

Usage

1. Add the ImprovedSpeechToText script to a GameObject in your scene.

2. Configure the script in the Inspector:

   • Select a microphone from the available list
   • Adjust the silence threshold, minimum speech duration, and maximum silence duration as needed

3. Run the scene. The script will automatically start listening and processing speech.

4. Speech recognition results will be logged to the Console. You can modify the ProcessRecognitionResult method to handle the results as needed for your application.

Customization

• To use a different Vosk model, change the modelName variable in the ImprovedSpeechToText script.
• Adjust the silenceThreshold, minSpeechDuration, and maxSilenceDuration parameters to fine-tune speech detection.

Troubleshooting

• If you encounter issues with library loading, check the Console for error messages from the VoskLoader script.
• Ensure that the Vosk model is correctly placed in the StreamingAssets folder.
• Verify that your microphone is properly connected and recognized by your system.

License

[MIT License]

Acknowledgements

This project uses the Vosk Speech Recognition Toolkit, which is distributed under the Apache 2.0 license.