Parakeet Models

Overview

OpenWhispr supports NVIDIA Parakeet models for local transcription via the sherpa-onnx runtime. Parakeet models offer:

Fast inference with INT8 quantization
Multilingual support (25+ languages)
Automatic language detection
Privacy-first local processing

Model Technology

runtime

string

sherpa-onnx (ONNX Runtime for cross-platform inference)

quantization

string

INT8 (8-bit integer quantization for efficient CPU inference)

framework

string

NVIDIA NeMo Transducer (streaming ASR architecture)

Storage Location

Models are cached locally at:

~/.cache/openwhispr/parakeet-models/

Each model directory contains:

encoder.int8.onnx
decoder.int8.onnx
joiner.int8.onnx
tokens.txt

Available Models

Parakeet TDT 0.6B v3 (680MB) — Recommended

name

string

Parakeet TDT 0.6B

modelId

string

parakeet-tdt-0.6b-v3

size

string

680MB

language

string

Multilingual (automatic detection)

supportedLanguages

array

25 languages including:

Germanic: English, German, Dutch, Swedish, Danish, Norwegian
Romance: French, Spanish, Portuguese, Italian, Romanian
Slavic: Russian, Ukrainian, Polish, Czech, Slovak, Bulgarian, Croatian, Slovenian
Baltic: Latvian, Lithuanian, Estonian
Other: Greek, Hungarian, Maltese, Finnish

downloadUrl

string

https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-parakeet-tdt-0.6b-v3-int8.tar.bz2

extractDir

string

sherpa-onnx-nemo-parakeet-tdt-0.6b-v3-int8

recommended

boolean

true

Supported Languages (Full List)

View all 25 supported languages

Code	Language	Code	Language
bg	Bulgarian	it	Italian
hr	Croatian	lv	Latvian
cs	Czech	lt	Lithuanian
da	Danish	mt	Maltese
nl	Dutch	pl	Polish
en	English	pt	Portuguese
et	Estonian	ro	Romanian
fi	Finnish	sk	Slovak
fr	French	sl	Slovenian
de	German	es	Spanish
el	Greek	sv	Swedish
hu	Hungarian	ru	Russian
—	—	uk	Ukrainian

When to Use Parakeet vs Whisper

Parakeet vs Whisper Comparison

Use Parakeet When:

You need fast multilingual transcription (25 European languages)
You want automatic language detection
You’re transcribing European languages (especially Slavic/Baltic)
You prefer lighter resource usage

Use Whisper When:

You need support for 58+ languages including Asian languages
You require highest accuracy for English
You’re working with technical/medical terminology
You need fine-grained model size options (6 models vs 1)

Downloading Models

Via UI

Models can be downloaded from Settings → Models → Parakeet Models.

Programmatically

const result = await window.api.downloadParakeetModel('parakeet-tdt-0.6b-v3', (progress) => {
  if (progress.type === 'progress') {
    console.log(`Downloading: ${progress.percentage}%`);
  } else if (progress.type === 'installing') {
    console.log('Extracting archive...');
  } else if (progress.type === 'complete') {
    console.log('Installation complete!');
  }
});

if (result.success) {
  console.log(`Model installed to: ${result.path}`);
}

Parakeet models are distributed as .tar.bz2 archives and are automatically extracted after download. The extraction process requires the tar command to be available on your system.

Checking Model Status

const status = await window.api.checkParakeetModelStatus('parakeet-tdt-0.6b-v3');

if (status.downloaded) {
  console.log(`Model path: ${status.path}`);
  console.log(`Size: ${status.size_mb}MB`);
} else {
  console.log('Model not downloaded');
}

Listing Models

const { models, cache_dir } = await window.api.listParakeetModels();

models.forEach(model => {
  console.log(`${model.model}: ${model.downloaded ? 'Downloaded' : 'Not downloaded'}`);
});

Deleting Models

const result = await window.api.deleteParakeetModel('parakeet-tdt-0.6b-v3');

if (result.deleted) {
  console.log(`Freed ${result.freed_mb}MB of disk space`);
}

Transcription

const audioBlob = /* ArrayBuffer from MediaRecorder */;

const result = await window.api.transcribeLocalParakeet(audioBlob, {
  model: 'parakeet-tdt-0.6b-v3',
  language: 'auto' // Automatic detection
});

if (result.success) {
  console.log(`Transcribed text: ${result.text}`);
} else {
  console.error(`Error: ${result.message}`);
}

Server Management

Starting the Server

const result = await window.api.startParakeetServer('parakeet-tdt-0.6b-v3');

if (result.success) {
  console.log(`Server running (this message won't appear - server starts automatically)`);
} else {
  console.error(result.reason);
}

Checking Server Status

const status = await window.api.getParakeetServerStatus();

console.log(`Ready: ${status.ready}`);
console.log(`Current model: ${status.currentModel}`);

Stopping the Server

await window.api.stopParakeetServer();

Server Pre-warming

Like Whisper, Parakeet supports server pre-warming on startup:

// Main process - automatic on startup
await parakeetManager.initializeAtStartup({
  localTranscriptionProvider: 'nvidia',
  parakeetModel: 'parakeet-tdt-0.6b-v3'
});

When LOCAL_TRANSCRIPTION_PROVIDER=nvidia is set in .env, the Parakeet server automatically starts on app launch, eliminating cold-start delays.

Provider Configuration

Set Parakeet as your transcription provider:

// Via settings store
import { getSettings, saveSettings } from '@/stores/settingsStore';

await saveSettings({
  localTranscriptionProvider: 'nvidia',
  parakeetModel: 'parakeet-tdt-0.6b-v3'
});

// This automatically saves to .env via saveAllKeysToEnvFile()

Binary Location

The sherpa-onnx binary is bundled with the app:

resources/bin/sherpa-onnx-{platform}-{arch}/

Supported platforms:

macOS: arm64, x64
Linux: x64, arm64
Windows: x64

Installation Check

const status = await window.api.checkParakeetInstallation();

if (status.installed && status.working) {
  console.log(`sherpa-onnx binary: ${status.path}`);
} else {
  console.error('Parakeet not available');
}

Parakeet models require the tar command for extraction. On Windows, ensure you have tar available (included in Windows 10+ by default).

Diagnostics

const diagnostics = await window.api.getParakeetDiagnostics();

console.log('Platform:', diagnostics.platform);
console.log('sherpa-onnx:', diagnostics.sherpaOnnx);
console.log('Downloaded models:', diagnostics.models);

Model Registry

All Parakeet models are defined in src/models/modelRegistryData.json:

{
  "parakeetModels": {
    "parakeet-tdt-0.6b-v3": {
      "name": "Parakeet TDT 0.6B",
      "description": "Fast multilingual ASR with auto language detection (25 languages)",
      "size": "680MB",
      "sizeMb": 680,
      "language": "multilingual",
      "supportedLanguages": ["bg", "hr", "cs", ...],
      "recommended": true,
      "downloadUrl": "https://github.com/k2-fsa/sherpa-onnx/releases/...",
      "extractDir": "sherpa-onnx-nemo-parakeet-tdt-0.6b-v3-int8"
    }
  }
}

IPC API

Models

Overview

Model Technology

Storage Location

Available Models

Parakeet TDT 0.6B v3 (680MB) — Recommended

Supported Languages (Full List)

When to Use Parakeet vs Whisper

Use Parakeet When:

Use Whisper When:

Downloading Models

Via UI

Programmatically

Checking Model Status

Listing Models

Deleting Models

Transcription

Server Management

Starting the Server

Checking Server Status

Stopping the Server

Server Pre-warming

Provider Configuration

Binary Location

Installation Check

Diagnostics

Model Registry

Build docs developers (and LLMs) love

IPC API

Models

​Overview

​Model Technology

​Storage Location

​Available Models

​Parakeet TDT 0.6B v3 (680MB) — Recommended

​Supported Languages (Full List)

​When to Use Parakeet vs Whisper

​Use Parakeet When:

​Use Whisper When:

​Downloading Models

​Via UI

​Programmatically

​Checking Model Status

​Listing Models

​Deleting Models

​Transcription

​Server Management

​Starting the Server

​Checking Server Status

​Stopping the Server

​Server Pre-warming

​Provider Configuration

​Binary Location

​Installation Check

​Diagnostics

​Model Registry

Build docs developers (and LLMs) love

Overview

Model Technology

Storage Location

Available Models

Parakeet TDT 0.6B v3 (680MB) — Recommended

Supported Languages (Full List)

When to Use Parakeet vs Whisper

Use Parakeet When:

Use Whisper When:

Downloading Models

Via UI

Programmatically

Checking Model Status

Listing Models

Deleting Models

Transcription

Server Management

Starting the Server

Checking Server Status

Stopping the Server

Server Pre-warming

Provider Configuration

Binary Location

Installation Check

Diagnostics

Model Registry