grabbou/parakeet-tdt-0.6b-v3-CoreAI

Parakeet TDT 0.6B v3 for Core AI

nvidia/parakeet-tdt-0.6b-v3 exported to Apple Core AI with the apple/coreai-models recipe:

uv run models/parakeet/export.py --streaming --dtype float16

Folder	Contents	Size	Platforms
streaming/	CoreAISpeech bundle: encoder, decoder step and joint, float16, 150-frame streaming window	1.2 GB	iOS, macOS

The streaming export slides a fixed window over audio of any length, so it transcribes both files and live audio. It recognizes 25 European languages and does not detect the language.

Use

With @react-native-ai/core-ai and the AI SDK, download streaming/ into app storage, then:

import { coreAI } from '@react-native-ai/core-ai'
import { experimental_transcribe as transcribe } from 'ai'

const model = coreAI.transcriptionModel({
  id: 'parakeet-tdt-0.6b-v3',
  source: { type: 'file', uri: pathToStreamingFolder },
})

const { text, segments } = await transcribe({ model, audio })

In Swift, load the folder with SpeechRecognitionModel(resourcesAt:) from the CoreAISpeech product of coreai-models.

Tested on iPhone 17 Pro with iOS 27.

License

The weights are NVIDIA's, under CC BY 4.0. This repository only converts them to Core AI format.