KokoroSharp

To perform basic text-to-speech synthesis using the CPU package, follow these steps:

1. Load the model using KokoroTTS.LoadModel(). This will automatically download the model (~320MB) if it is not present.
2. Retrieve a voice using KokoroVoiceManager.GetVoice(string voiceName).
3. Use tts.SpeakFast(string text, KokoroVoice voice) to synthesize and play the text.

Language detection is automated based on the capabilities of the selected voice.

KokoroTTS tts = KokoroTTS.LoadModel(); // Load or download the model
KokoroVoice heartVoice = KokoroVoiceManager.GetVoice("af_heart"); 

// Speak text from console input
while (true) 
{
    string input = Console.ReadLine();
    tts.SpeakFast(input, heartVoice);
}

When using the KokoroSharp.Android package, you must call KokoroSharpAndroid.Init() before starting the standard workflow. This initialization step also handles the synchronous downloading of voices.

KokoroSharpAndroid.Init();
// Proceed with standard KokoroTTS workflow

To use NVIDIA GPU acceleration via CUDA, follow these steps:

1. Install Packages: Choose KokoroSharp.GPU.Windows, KokoroSharp.GPU.Linux, or KokoroSharp.GPU via NuGet.
2. Install Dependencies:
   • Download and install CUDA Toolkit (v12.x).
   • Download and install cuDNN.
   • Ensure both are added to your SYSTEM PATH.
3. Restart: Restart your IDE or terminal to sync the system path.
4. Configure Code: Use SessionOptions to append the CUDA execution provider when loading the model.

var options = new SessionOptions();
options.AppendExecutionProvider_CUDA();
KokoroTTS tts = KokoroTTS.LoadModel(sessionOptions: options);
tts.SpeakFast("Hello from GPU!", KokoroVoiceManager.GetVoice("af_heart"));

KokoroSharp can be installed on Windows, Linux, and MacOS using the NuGet Package Manager or the .NET CLI.

• For CPU-based inference (plug-and-play), use the KokoroSharp.CPU package.
• For GPU support, refer to the RUNNING_ON_GPU.md guide in the repository.
• For Android, use the KokoroSharp.Android package.
• For iOS, support is available via the standard package, utilizing a custom CrossPlatformHelper.CustomAudioPlayer for playback.

# Example: Install via .NET CLI
dotnet add package KokoroSharp.CPU

KokoroSharp provides different packages depending on your target platform and whether you want CPU or GPU acceleration:

While there is no dedicated KokoroSharp package for CoreML, you can use it if you build and install the CoreML ONNX runtime manually.

1. Install the base KokoroSharp package via NuGet.
2. Build and install the CoreML ONNX runtime on your system.
3. Use options.AppendExecutionProvider_CoreML() when loading the model.

var options = new SessionOptions();
options.AppendExecutionProvider_CoreML();
KokoroTTS tts = KokoroTTS.LoadModel(sessionOptions: options);
tts.SpeakFast("Hello from GPU!", KokoroVoiceManager.GetVoice("af_heart"));

If you encounter errors during execution, check the following:

• EntryPointNotFoundException:
  • You likely have the wrong KokoroSharp package installed.
  • You might have multiple conflicting runtime packages installed. Delete excess runtimes to resolve this.
• DllNotFoundException:
  • You have no runtime installed. For a simple, plug-and-play experience, use the KokoroSharp.CPU package.

Note: Always restart your IDE or terminal after installing system-level dependencies (like CUDA or cuDNN) to ensure the SYSTEM PATH is updated.

To use the Chinese-specific model and voices, load the model using the KModel.zh_float32 constant. The v1.1-zh voices are bundled in the package under voices/voices-zh and are loaded automatically.

KokoroTTS tts = KokoroTTS.LoadModel(KModel.zh_float32); 
tts.SpeakFast("你好，世界！", KokoroVoiceManager.GetVoice("zf_001"));

If you need to generate raw audio data instead of playing it immediately, use the KokoroWavSynthesizer class.

• LoadModel(): Loads the synthesizer model.
• Synthesize(string text, KokoroVoice voice): Returns the audio as a byte[]. An async version is also available.
• SynthesizeWithTimestampsAsync(string text, KokoroVoice voice): Returns audio with per-phoneme timestamps, useful for lip-syncing.
• SaveAudioToFile(byte[] audioBytes, string filePath): A static helper to save the byte array to a file.

KokoroWavSynthesizer synth = KokoroWavSynthesizer.LoadModel();
byte[] audioBytes = synth.Synthesize("Hello world!", KokoroVoiceManager.GetVoice("af_heart"));

// Save the result to a file
KokoroWavSynthesizer.SaveAudioToFile(audioBytes, "hello.wav");