Speech

Generate speech on device with a text-to-speech GGUF model and a codec.cpp vocoder, which turns the model’s audio tokens into sound. Download both:

import { downloadModel, llama } from '@react-native-ai/llama'
import { generateSpeech } from 'ai'

const modelPath = await downloadModel(
  'OuteAI/OuteTTS-0.3-500M-GGUF/OuteTTS-0.3-500M-Q4_K_M.gguf'
)
const vocoderPath = await downloadModel(
  'BricksDisplay/codec.cpp-gguf/wavtokenizer-large-speech-75tokens.gguf'
)

const model = llama.speechModel(modelPath, { vocoderPath })
await model.prepare()

const { audio } = await generateSpeech({
  model,
  text: 'Hello from on-device speech!',
})

console.log(audio.uint8Array)

The audio is 16-bit mono WAV at the vocoder’s sample rate. Only text is used: voice, speed, and language are ignored. See the API reference for the model options.