Live transcription · Multilingual

Parakeet TDT 0.6B v3

NVIDIA's multilingual speed flagship — 25 European languages, TDT transducer with native word timestamps

Multilingual · 7.8× · 876.6 MB

Install in OpenASR Desktop

Opens the desktop app and downloads this model for you.

Don't have the app yet? Download OpenASR Desktop

On-deviceOn this computer
interview.m4a12:04

let's line up this week's release plan,

the beta build can ship on Friday,

then loop in design and QA,

Why it fits

Parakeet TDT 0.6B v3 is NVIDIA's 600M-parameter multilingual speech-recognition model, extending the English-only v2 to 25 European languages using the Granary multilingual corpus (~660k hours of pseudo-labeled speech plus ~10k hours of human transcriptions). Architecturally it pairs a FastConformer encoder with a Token-and-Duration Transducer (TDT) decoder: each decode step predicts both the next token and how many 80 ms frames it occupies, then skips ahead by that duration — which is what makes the family a speed benchmark for local transcription, and what gives it model-native word timestamps. The model auto-detects the spoken language, preserves punctuation and capitalization, and is released under the permissive CC-BY-4.0 license. This OpenASR repo repackages the weights as .oasr packs that run natively in the OpenASR runtime — no Python at inference time, all decoding local. The q8_0 build is the recommended default; q4_k is the smallest build for tight-memory devices and fp16 is for maximum fidelity or verification.

What you can do with it

  • 🇪🇺 25 European languages — one 0.6B checkpoint covers bg/hr/cs/da/nl/en/et/fi/fr/de/el/hu/it/lv/lt/mt/pl/pt/ro/sk/sl/es/sv/ru/uk, auto-detecting the spoken language with no prompting
  • Built for throughput — the Token-and-Duration Transducer (TDT) decoder predicts how many frames each token spans and skips ahead, cutting decode steps versus conventional transducers
  • 🕑 Native word timestamps — the duration head is the model's own alignment output, so word-level timings come from the model rather than a uniform approximation
  • ✍️ Punctuation and capitalization — trained on transcripts that preserve both (Granary + NeMo ASR Set 3.0; ~670k hours), so the raw decode reads like text

Other models

Developer details CLI commands, file hashes, and per-quant downloads — for scripting and verification.

Downloads by quantization

Pull stringSizeQuant
parakeet-tdt-0.6b-v3:fp16 1.3 GB fp16
open .oasr
parakeet-tdt-0.6b-v3:q8default 876.6 MB q8_0
open .oasr
parakeet-tdt-0.6b-v3:q4 624.2 MB q4_k
open .oasr

Model facts

Size876.6 MB
Speed7.8× real-time
LanguageMultilingual
Familyparakeet-tdt
Vendornvidia
Released2025-08-14
Peak memory1 GB
Quantizationq8_0
LicenseCC-BY-4.0
Revision4a16094df4ec3ab6081089a5aff415c2042342b7
sha2564416134e5f4d42690ecfe6b2be8c30fcb3186eb5491434195355a1597f497454

Usage

These are CLI / local-server examples. The desktop app runs this model without typing a command — use Install above.

bash · transcribe a file
$ openasr pull parakeet-tdt-0.6b-v3:q8
↓ parakeet-tdt-0.6b-v3.oasr  876.6 MB  ✓ verified sha256
$ openasr transcribe meeting.wav --backend native --model-pack ~/.openasr/models/parakeet-tdt-0.6b-v3/q8_0/parakeet-tdt-0.6b-v3-q8_0.oasr
✓ local transcript · 0 bytes sent
bash · serve a local API
$ openasr serve --backend native --model-pack ~/.openasr/models/parakeet-tdt-0.6b-v3/q8_0/parakeet-tdt-0.6b-v3-q8_0.oasr --addr 127.0.0.1:8080
▶ http://127.0.0.1:8080 · model=parakeet-tdt-0.6b-v3 · 0 bytes will leave this host
python · client.py
from openai import OpenAI
client = OpenAI(base_url="http://127.0.0.1:8080/v1", api_key="local")
audio = open("meeting.wav", "rb")
text = client.audio.transcriptions.create(model="parakeet-tdt-0.6b-v3", file=audio)