Parakeet RNNT 1.1B

Parakeet RNNT 1.1B

Transcribes English speech into lowercase text. A 1.1B FastConformer Transducer model from the NVIDIA NeMo and Suno.ai teams, for developers who add speech recognition to apps.

Made by nvidia on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Parakeet RNNT 1.1B turns spoken English into written text. NVIDIA NeMo and Suno.ai built it together as an XXL FastConformer Transducer with about 1.1B parameters. It outputs text in the lowercase English alphabet and runs through the NeMo toolkit or Hugging Face Transformers.

Who it suits: It suits developers and researchers who need English speech to text on their own machines, either with the checkpoint as it is or as a base for fine-tuning.

Industry | Who uses it | How they use it | Example
Media | Podcast and video teams | Transcribe English recordings into text | A folder of interview recordings turned into text
Customer service | Support teams | Turn English call audio into text for search or review | A recorded support call transcribed
Software | App developers | Add English speech to text to a product through NeMo or Transformers | A voice note feature in an app
Research | Speech researchers | Fine-tune the checkpoint on another dataset | Adapting the model to a specialist vocabulary

You can skip it if your audio is not in English, since the card lists English only.

Where to use it:

• The NeMo toolkit, installed after the latest PyTorch
• Hugging Face Transformers, from version 5.13.0
• A command line, using the transcription script from the NeMo repository
• Windows, macOS or Linux with Python, to download the files

Use case | Who | What they get back
Transcribe one file | Developers | The spoken words as lowercase text
Transcribe a folder | Data teams | A transcript for each audio file in the folder
Fine-tune on your data | Researchers | A model adapted to another dataset

How to set it up:

• Python
• The latest PyTorch, installed before NeMo
• NeMo, or Transformers 5.13.0 or newer
• Room for about 4.3 GB of weights in F32, or about 535 MB at 4-bit by estimate

1. Install the latest PyTorch.
2. Install the NeMo toolkit.
3. Load the model by its name in Python, as shown below.

pip install nemo_toolkit['all']

import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.EncDecRNNTBPEModel.from_pretrained(model_name="nvidia/parakeet-rnnt-1.1b")

To fetch the files yourself first:

pip install -U huggingface_hub
hf download nvidia/parakeet-rnnt-1.1b

How to use it: Pass a list of audio files to the transcribe call and print the first result. The file name below is the card's sample, so swap in your own English recording.

output = asr_model.transcribe(['2086-149220-0033.wav'])
print(output[0].text)

For a whole folder, run the NeMo script with the model name and the folder of audio files.

python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py
 pretrained_name="nvidia/parakeet-rnnt-1.1b"
 audio_dir="<DIRECTORY CONTAINING AUDIO FILES>"

To use Transformers instead, install version 5.13.0 or newer.

pip install "transformers>=5.13.0"

Tool | What it does
from_pretrained | Loads the model by its name
asr_model.transcribe | Turns a list of audio files into text, one result per file
transcribe_speech.py | A NeMo script that transcribes every audio file in a folder
AutoModelForRNNT | Loads the model when you use Transformers

Example: Give it a short English WAV recording and print the first result's text. You get the spoken words back as lowercase text.

For newcomers and experts: It listens to an English recording and writes down what was said. For a first try, install NeMo after PyTorch, load the model with the Python above, and transcribe one short English WAV file.

Underneath, it is a FastConformer Transducer in its XXL size, about 1.1B parameters, built jointly by the NVIDIA NeMo and Suno.ai teams. You can use the checkpoint for inference as it is, or fine-tune it on another dataset. The Transformers route needs version 5.13.0 or newer.

English audio file goes in
NeMo or Transformers loads the checkpoint
The FastConformer Transducer decodes the speech
Lowercase English text comes out

Good to know: The output uses the lowercase English alphabet only, so expect no capital letters. Downloading the public release counts as accepting its license terms.

• License: CC BY 4.0
• Cost to run: no API key or paid service is named
• Data it reaches: reads the audio files you give it and downloads model files from Hugging Face, and the card's sample first downloads an example WAV file from the web
• Not for: languages other than English, since the card lists English only

On Hugging Face, 13 November 2026

• Task: automatic speech recognition
• Library: nemo
• Parameters: 1.1B
• Weights: about 4.3 GB in F32; about 535 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 8,709
• Likes: 189
• License: cc-by-4.0
• Languages: en
• Quantized versions: 14
• Fine-tunes: 2
• Tags: nemo, safetensors, parakeet_rnnt, feature-extraction, automatic-speech-recognition, speech, audio, transducer, fastconformer, conformer, pytorch, hf-asr-leaderboard

Links

• Model: https://huggingface.co/nvidia/parakeet-rnnt-1.1b
• Files: https://huggingface.co/nvidia/parakeet-rnnt-1.1b/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download nvidia/parakeet-rnnt-1.1b

Then follow the usage in its model card. Nothing in this kit runs by itself.