Models / NVIDIA
Audio

NVIDIA Parakeet TDT 0.6B V3 Realtime

Real-time multilingual transcription across EU official languages with automatic language detection

Try now
read docs

About model

Parakeet TDT 0.6B V3 Realtime is the live-transcription version of NVIDIA's 600-million-parameter multilingual speech-to-text model, serving the same FastConformer-TDT architecture for streaming audio. Trained on NVIDIA's Granary dataset (670,000+ hours of audio), the model automatically detects the input language across EU official languages and transcribes as audio arrives, returning text with punctuation, capitalization, and timestamps. It carries the accuracy that places Parakeet TDT 0.6B V3 among the highest throughput multilingual models on the HuggingFace Open ASR Leaderboard, with a 6.34% average word error rate. Available on Together AI.

‍

Average Word Error Rate

6.34%

Accuracy on the HuggingFace Open ASR Leaderboard

Hours of Training Audio

670K+

Trained on NVIDIA's Granary multilingual dataset

Language Detection

Automatic

Transcribes EU official languages without prompting

Model key capabilities
  • Live Multilingual Transcription: Streaming transcription across EU official languages with automatic language detection and no additional prompting
  • Real-Time Architecture: FastConformer-TDT design built for high-throughput inference on continuous audio
  • Production-Ready Output: Automatic punctuation, capitalization, and word-level timestamps with every transcription
  • Production-Ready Infrastructure: 99.9% SLA, available on serverless and dedicated infrastructure

  • API usage

    • cURL
    • Python
    • Typescript

    Endpoint:

    nvidia/parakeet-tdt-0.6b-v3-realtime

    curl -X POST "https://api.together.xyz/v1/audio/transcriptions" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -F "model=nvidia/parakeet-tdt-0.6b-v3-realtime" \
  -F "language=en" \
  -F "response_format=json" \
  -F "timestamp_granularities=segment"

    from together import Together

client = Together()
response = client.audio.transcribe(
    model="nvidia/parakeet-tdt-0.6b-v3-realtime",
    language="en",
    response_format="json",
    timestamp_granularities="segment"
)
print(response.text)

    import Together from "together-ai";

const together = new Together();

const response = await together.audio.transcriptions.create(
  model: "nvidia/parakeet-tdt-0.6b-v3-realtime",
  language: "en",
  response_format: "json",
  timestamp_granularities: "segment"
});
console.log(response)

  • Model card

    Architecture Overview:
    • FastConformer-TDT (Token-and-Duration Transducer) architecture with 600 million parameters
    • Automatic language detection and transcription across EU official languages
    • Streaming output with punctuation, capitalization, and timestamps

    Training Methodology:
    • Trained on NVIDIA's Granary multilingual corpus with 670,000+ hours of audio
    • Stage 2 fine-tuning on approximately 7,500 hours of high-quality, human-transcribed data

    Performance Characteristics:
    • 6.34% average word error rate on the HuggingFace Open ASR Leaderboard
    • Among the highest throughput multilingual models on the leaderboard

    ‍

  • Prompting

    Together AI API Access:
    • Access Parakeet TDT 0.6B V3 Realtime via Together AI APIs using the endpoint nvidia/parakeet-tdt-0.6b-v3-realtime
    • Authenticate using your Together AI API key in request headers
    • Stream 16kHz monochannel audio and receive live transcriptions with punctuation and timestamps
    • The model automatically detects the language of the input audio, so no language specification is required
    • Available on Together AI serverless and dedicated infrastructure

    ‍

  • Applications & use cases

    Live Voice Agents:
    • Transcribe caller speech in real time as the listening layer for voice agents built on the Together endpoint
    • Route multilingual callers without a separate language-identification step, using built-in automatic detection
    • Feed punctuated, capitalized transcripts directly into downstream LLM reasoning

    Live Captioning & Accessibility:
    • Caption meetings, webinars, and broadcasts as they happen across EU official languages
    • Generate word-level timestamps for synchronized on-screen captions
    • Support accessibility workflows that require low-delay, readable transcripts

    Multilingual Customer Service:
    • Transcribe live support calls across European markets from a single endpoint
    • Power real-time agent-assist tools that surface answers while the customer is speaking
    • Capture clean transcripts for quality review without a post-processing pass

    ‍

Related models
new
Transcribe
NVIDIA Nemotron 3 ASR Streaming 0.6B

  • Price/min

    $0.0015
  • Parameters
    600M
No items found.
Audio
new
Transcribe
NVIDIA Parakeet TDT 0.6B v3

  • Price/min

    $0.0015
No items found.
Transcribe
new
Transcribe
NVIDIA Nemotron 3.5 ASR

  • Price/min

    $0.0045/min
  • Parameters
    0.6B
No items found.
Transcribe
Model specifications
Model data
  • Model provider
    NVIDIA
  • Type
    Audio
  • Deployment
    Serverless
    On-Demand Dedicated
  • Endpoint
    nvidia/parakeet-tdt-0.6b-v3-realtime
  • Parameters
    600M
  • Price

    $0.0035 / min

  • Input modalities
    Audio
  • Output modalities
    Text
  • Category
    Transcribe
Run in Playground
Quickstart docs
Deploy model