parakeet-ultra

Its sibling Parakeet Redux is the ternary version of the same architecture, 178 MB, built for CPUs and Apple silicon. Read the release post for the story behind both models.

Usage

Run it with Photon. Every accuracy number on this page is Photon on an NVIDIA GPU.

# pip install moondream
import moondream as md

with md.photon("moondream/parakeet-ultra") as speech:
    result = speech.transcribe(audio="speech.wav")
    print(result["text"])

    # timestamps="segment": one entry per sentence, with its start
    # and end in seconds; "word" adds the start and end of every word
    result = speech.transcribe(audio="speech.wav", timestamps="word")
    for segment in result["segments"]:
        print(segment["start"], segment["end"], segment["text"])
        for word in segment["words"]:
            print(word["start"], word["end"], word["word"])

Performance

Real-time factor: seconds of audio transcribed per second of wall clock, higher is faster. One NVIDIA B200, 128 requests in flight, the same files on both sides; NeMo 3.0 runs the original checkpoint with its own transcribe call at batch 128, Photon runs parakeet-ultra.

Benchmarks

Word error rate in percent, lower is better. Both models are scored on the same files with the Open ASR Leaderboard's own pipeline as of September 2026 (its normalizers and compound-merging alignment, with the FLEURS references prepared as the leaderboard's text column is). parakeet-ultra runs in Photon on an NVIDIA GPU, the original in NeMo in bf16.

Open ASR Leaderboard

The seven English test sets of the Hugging Face Open ASR Leaderboard: audiobooks (LibriSpeech), meetings (AMI), earnings calls (Earnings-22), podcasts and YouTube (GigaSpeech), financial calls (SPGISpeech) and parliament (VoxPopuli).

FLEURS

The FLEURS test split: read Wikipedia sentences, a few hundred per language, for all 25 languages the model supports.

Business speech (AA-WER style)

The AMI and VoxPopuli test sets with the Artificial Analysis cleaning applied, and Earnings-22 scored in 30-second chunks joined per call, the way the AA-WER benchmark does it.

Background noise

The clean sets with MUSAN noise (the half of the corpus not used for training) mixed in at a fixed signal-to-noise ratio; 0 dB means the noise is as loud as the speech.

Long-form

Eleven complete TED-LIUM 3 talks of 10-20 minutes each. parakeet-ultra transcribes them through Photon, whose segmenter cuts each talk at pauses found by the model's VAD head into segments of at most 30 seconds; the original runs NeMo's own long-audio path.

Notes

• Based on parakeet-tdt-0.6b-v3 by NVIDIA. Languages, tokenizer and output conventions (punctuation, casing, numerals) are the original's.
• Runs with Photon.
• Long audio is segmented by the model itself: the weights carry a small voice-activity head on the encoder's