SpeechSynthesizer

classin packageklyn.genai.tts

public class SpeechSynthesizer implements AutoClosable

SpeechSynthesizer

All Implemented Interfaces: AutoClosable

Resident Qwen3-TTS CustomVoice synthesis from GGUF or original HF Safetensors, implemented in Klyn/C++ using the existing Klyn NN backend. No additional inference library, Python service or external executable is used. Synthesis produces typed PCM, never implicit playback or microphone activation. Auto selects a compatible NVIDIA GPU, otherwise CPU. Packed projections execute on that device; causal audio kernels use synchronized CPU storage. An explicit GPU never falls back. NPU execution is not available yet. A synthesizer is single-owner; only SpeechStream.cancel is cross-thread.

Example

try speech = SpeechSynthesizer.load("sakasegawa/Qwen3-TTS-12Hz-0.6B-CustomVoice-GGUF"): speech.synthesize("Hello", voice="ryan", language="en").play()

Properties

Modifier and Type	Member	Description
public readonly property	device `device as Device = Device.cpu`	Actual device, not the original selector.
public readonly property	sampleRate `sampleRate as Int`	Output frequency in Hz.
public readonly property	voices `voices as String[]`	Fresh array of built-in voice names.

Constructors

Modifier and Type	Member	Description
public	SpeechSynthesizer `SpeechSynthesizer(notnull folder as FolderPath, notnull device as Device = Device.auto, maxContext as Long = 1024l, maxFrames as Long = 512l, seed as ULong = 0ul, verbose as Boolean = false)`	Opens original HF Safetensors with the nested speech_tokenizer codec.
public	SpeechSynthesizer `SpeechSynthesizer(notnull path as FilePath, notnull device as Device = Device.auto, maxContext as Long = 1024l, maxFrames as Long = 512l, seed as ULong = 0ul, verbose as Boolean = false)`	Opens a local CustomVoice GGUF and keeps its parameters resident.

Methods

Modifier and Type	Member	Description
public override	close `close() as Void`	Cancels and joins the active worker, then releases model parameters.
public static	load `load(notnull repository as String, quantization as ModelQuantization = ModelQuantization.AUTO, notnull device as Device = Device.auto, maxContext as Long = 1024l, maxFrames as Long = 512l, seed as ULong = 0ul, revision as String = "main", offline as Boolean = false, verbose as Boolean = false, token as String = "", format as ModelFormat = ModelFormat.AUTO) as SpeechSynthesizer`	Resolves a pinned Hub quantization through the shared verified cache, then loads it.
public	stream `stream(notnull text as String, notnull voice as String = "", notnull language as String = "auto", notnull instructions as String = "") as SpeechStream`	Starts bounded, cancellable incremental synthesis on a registered Klyn worker.
public	synthesize `synthesize(notnull text as String, notnull voice as String = "", notnull language as String = "auto", notnull instructions as String = "") as AudioClip`	Produces a complete independent mono Float32 clip without playback.

Inherited Methods

methodInherited Methods from AutoClosable: close