Assembling...
Generative speech model using flow matching to perform text-to-speech, noise removal, cross-lingual style transfer, and content editing. Produces natural speech by learning from diverse unstructured audio without explicit labels.
First generative model to perform in-context text-to-speech across six languages
Flow matching architecture generates speech 20x faster than autoregressive models
Style transfer preserves speaker identity across language boundaries
Generate natural-sounding narration for long-form content with consistent voice quality.
Deliver voice alerts and notifications with expressive, human-like speech synthesis.
Produce audio content in multiple languages from a single text source.
Power low-latency voice responses in interactive applications and games.
// Voicebox — Text-to-Speech
import { synthesize } from "@arkitekton/voice";
const audio = await synthesize({
model: "vm-meta-001",
vendor: "meta",
input: "Hello, welcome to Arkitekton.",
voice: "alloy",
response_format: "mp3",
speed: 1.0,
});
// Play the audio
const blob = new Blob([audio], { type: "audio/mp3" });
const url = URL.createObjectURL(blob);
const player = new Audio(url);
player.play();Open research in generative speech and massively multilingual models