whatmodel.sh / stt / Moonshine Streaming Small

Moonshine Streaming Small

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Moonshine Streaming Small is an open-weights speech-to-text model from Useful Sensors (Moonshine AI), released January 2026. Its published size is 0.14B parameters. Licence: mit.

Catalogue identity notes

Exact checkpoint moonshine-ai/moonshine-streaming-small (CSV entry 'usefulsensors/moonshine-streaming-small'; the Hub redirects the old UsefulSensors org name). Params: HF safetensors total. released: HF repo creation month.

45.6±2.9#119 of 162 speech-to-text modelsfrom 4 benchmark results across 3 sources

Score within the Open ASR English short-form WER (March 2026) comparison set only.

$ how it compares

The speech-to-text models ranked either side of Moonshine Streaming Small, out of 176 with a score, then more from Useful Sensors (Moonshine AI).

#modelscorewhereRAMreleased
130ESPnet Multiconvformer (no LM) ESPnet46±4local~0.8GB2024-03
131STT En FastConformer-Transducer Large NVIDIA46±3local~1.1GB2023-06
132Moonshine Streaming Small Useful Sensors (Moonshine AI)46±3local~1.2GB2026-01
133SenseVoice Small Alibaba (FunAudioLLM)46±3local~0.3GB2024-07
134STT En FastConformer-CTC Large NVIDIA45±3local~1.1GB2023-06
more from Useful Sensors (Moonshine AI)
108Moonshine Streaming Medium Useful Sensors (Moonshine AI)50±3local~1.8GB2026-01
155Moonshine Base Useful Sensors (Moonshine AI)38±3local~0.8GB2024-10
159Moonshine Tiny Useful Sensors (Moonshine AI)29±3local~0.6GB2024-10

see all speech-to-text models · find the ones that fit your machine

$ what it's good at

other tests4 not in a task · 97% of its score
testresultvs the others in its rundifficultycountswhy
LibriSpeech test-clean WER2.5% ✎▁···▁▁▂·▄▄██ beats 16/56avg 2.3%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 38%blunt
Open ASR English short-form WER (March 2026)7.8% ✎▁▁▁▃··▁▁▂██▆ beats 39/86avg 8.2%▇▇▇▇▇▇▇▇▇▇▇▇ 30%blunt
FLEURS English test WER8.6% ✎▁··▁▁▁▁▃▁▃█▁ beats 7/56avg 5.8%▇▇▇▇▇▇ 15%blunt
Picovoice STT benchmark · English streaming 4-set mean WER13.4% ✎█·█···▄▄▄··▄ beats 4/7avg 8.3%▇▇▇▇▇▇ 15%typical

$ where to run it

Other ways to run it

  1. Run it as published · needs ~1.17 GB · from huggingface.co/moonshine-ai (official)
    uv run --with transformers --with torch --with librosa python -c "import librosa; from transformers import MoonshineStreamingForConditionalGeneration as M, AutoProcessor as P; r='moonshine-ai/moonshine-streaming-small'; m=M.from_pretrained(r); p=P.from_pretrained(r); i=p(librosa.load('audio.wav', sr=16000)[0], return_tensors='pt', sampling_rate=16000); print(p.decode(m.generate(**i, max_length=int(i.attention_mask.sum()*6.5/16000))[0], skip_special_tokens=True))"

Official weights and documentation: huggingface.co/moonshine-ai ↗

$ how the score was worked out · transcribe.cpp model catalog… 53%, Open ASR Leaderboard 30%, Picovoice speech-to-text-benc… 15%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
transcribe.cpp model catalog · LibriSpeech test-clean WER of validated full-precision ports2▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 53%MIT
Open ASR Leaderboard1▇▇▇▇▇▇▇▇▇ 30%Apache-2.0
Picovoice speech-to-text-benchmark · English batch-engine WER (README)1▇▇▇▇ 15%Apache-2.0
prior · pull toward the middle▇ 3%
weighted average= 45.6

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json