SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.12241 · 2026

Moonshine v2: Ergodic Streaming Encoder ASR for Latency-Critical Speech Applications

James Wang, Manjunath Kudlur, Pete Warden, Evan King

arXiv · PDF · Open in the Atlas

Code that ran

12 of 15 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 4 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 8 failures of our instrument; 3 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2602.12241.json · badge: /api/ran/2602.12241.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
moonshine-ai/moonshine canonical (classifier label)
classifier basis: census
6 ran · 2 no recorded run
SYSTRAN/faster-whisper canonical (classifier label)
classifier basis: census
6 ran · 1 no recorded run
FunctionStatusWhere it lives
collect_chunks Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
SYSTRAN/faster-whisper/faster_whisper/vad.py
code served (permissive licence) · get_code("7f192506a5620211")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a small audio array (sampling_rate * duration samples)
# Using 16000 Hz, 2 seconds of audio = 32000 samples
audio = np.random.randn(32000).astype(np.float32)

# Create a list of chunks with start and end indices
chunks = [
    {'start': 0, 'end': 8000},
    {'start': 8000, 'end': 16000},
    {'start': 16000, 'end': 24000},
    {'start': 24000, 'end': 32000},
]

sampling_rate = 16000
max_duration = 1.0  # 1 second

ARGS = {
    'audio': audio,
    'chunks': chunks,
    'sampling_rate': sampling_rate,
    'max_duration': max_duration,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
format_timestamp Ran
ran; our contract draft was wrong, not the code
SYSTRAN/faster-whisper/faster_whisper/utils.py
code served (permissive licence) · get_code("27a52b01d1242740")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

# Setup code for format_timestamp function

ARGS = {
    'seconds': 3661.456,
    'always_include_hours': False,
    'decimal_marker': '.'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_cache_dir Ran
ran; our contract draft was wrong, not the code
moonshine-ai/moonshine/language-bindings/python/src/moonshine_voice/download_file.py
pointer only (licence: NOASSERTION) · get_code("d216fc2ce888c4e8")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, moonshine-ai/moonshine (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_compression_ratio Ran
ran, honoured the contract we drafted
SYSTRAN/faster-whisper/faster_whisper/transcribe.py
code served (permissive licence) · get_code("df860301447ca3dd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import zlib

# The function takes a string and computes compression ratio
# It encodes the string to UTF-8 bytes, compresses it with zlib,
# and returns the ratio of original size to compressed size

ARGS = {
    "text": "The quick brown fox jumps over the lazy dog. " * 10
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_end Ran
ran, honoured the contract we drafted
SYSTRAN/faster-whisper/faster_whisper/utils.py
code served (permissive licence) · get_code("a11a18b0223db329")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# Construct a valid segments list for the get_end function
# The function expects List[dict] where each dict has 'words' (list of dicts with 'end' key) and 'end' key

segments = [
    {
        'words': [
            {'end': 1.5},
            {'end': 2.0}
        ],
        'end': 2.5
    },
    {
        'words': [
            {'end': 3.0},
            {'end': 3.5}
        ],
        'end': 4.0
    }
]

ARGS = {
    'segments': segments
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_logger Ran
ran; our contract draft was wrong, not the code
SYSTRAN/faster-whisper/benchmark/utils.py
code served (permissive licence) · get_code("ce80eafaf33fae23")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned RootLogger. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import logging
from typing import Optional

# The function get_logger takes an optional string name parameter
# Reading the body: it uses logging.getLogger(name), which accepts None or a string
# So we can test with both None and a string value

ARGS = {
    "name": None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type RootLogger — a RootLogger output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a RootLogger output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
hash_file Ran
ran; our contract draft was wrong, not the code
moonshine-ai/moonshine/language-bindings/python/src/moonshine_voice/download_file.py
pointer only (licence: NOASSERTION) · get_code("4dd63075dea3762c")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, moonshine-ai/moonshine (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

model_arch_to_string Ran
ran; our contract draft was wrong, not the code
moonshine-ai/moonshine/language-bindings/python/src/moonshine_voice/moonshine_api.py
pointer only (licence: NOASSERTION) · get_code("c0b8d2436a8e0deb")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, moonshine-ai/moonshine (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

moonshine_options_array Ran
ran; our contract draft was wrong, not the code
moonshine-ai/moonshine/language-bindings/python/src/moonshine_voice/moonshine_api.py
pointer only (licence: NOASSERTION) · get_code("9a27c4a8b45fda56")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, moonshine-ai/moonshine (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

pad_or_trim Ran
ran; our contract draft was wrong, not the code
SYSTRAN/faster-whisper/faster_whisper/audio.py
code served (permissive licence) · get_code("4db3da4d6c68fdfd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (128, 3000) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a small numpy array with shape (128, 2500) on the last axis
# This will test both padding and trimming scenarios
array = np.random.randn(128, 2500)

ARGS = {
    "array": array,
    "length": 3000,
    "axis": -1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (128, 3000) — shape (128, 3000) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
string_to_model_arch Ran
ran, honoured the contract we drafted
moonshine-ai/moonshine/language-bindings/python/src/moonshine_voice/moonshine_api.py
pointer only (licence: NOASSERTION) · get_code("efd406f04b8eba8f")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, moonshine-ai/moonshine (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

write_cached_embeddings_tsv Ran
ran, honoured the contract we drafted
moonshine-ai/moonshine/language-bindings/python/src/moonshine_voice/cached_embeddings.py
pointer only (licence: NOASSERTION) · get_code("0d4210a3940c2807")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, moonshine-ai/moonshine (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

crc32c_file Not yet run
unverified (harvested, no recorded run)
moonshine-ai/moonshine/language-bindings/python/src/moonshine_voice/download_file.py
pointer only (licence: NOASSERTION) · get_code("fab9f7d50af01a81")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, moonshine-ai/moonshine (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

download_model Not yet run
unverified (harvested, no recorded run)
SYSTRAN/faster-whisper/faster_whisper/utils.py
code served (permissive licence) · get_code("bd84f59797ba624b")
How it ran

no invocation recorded.

spoken_form Not yet run
unverified (harvested, no recorded run)
moonshine-ai/moonshine/language-bindings/python/src/moonshine_voice/alphanumeric_listener.py
pointer only (licence: NOASSERTION) · get_code("c1f6d9a28c7f13f2")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, moonshine-ai/moonshine (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 8 are also present in other papers whose code we hold (8 with at least one paper dated earlier, 0 the same month, 0 later) and 7 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
crc32c_file 1 other paper
1 dated earlier
2410.15608 (2024-10)
Moonshine: Speech Recognition for Live Transcription and Voice Commands
get_cache_dir 1 other paper
1 dated earlier
2410.15608 (2024-10)
Moonshine: Speech Recognition for Live Transcription and Voice Commands
hash_file 1 other paper
1 dated earlier
2410.15608 (2024-10)
Moonshine: Speech Recognition for Live Transcription and Voice Commands
model_arch_to_string 1 other paper
1 dated earlier
2410.15608 (2024-10)
Moonshine: Speech Recognition for Live Transcription and Voice Commands
moonshine_options_array 1 other paper
1 dated earlier
2410.15608 (2024-10)
Moonshine: Speech Recognition for Live Transcription and Voice Commands
spoken_form 1 other paper
1 dated earlier
2410.15608 (2024-10)
Moonshine: Speech Recognition for Live Transcription and Voice Commands
string_to_model_arch 1 other paper
1 dated earlier
2410.15608 (2024-10)
Moonshine: Speech Recognition for Live Transcription and Voice Commands
write_cached_embeddings_tsv 1 other paper
1 dated earlier
2410.15608 (2024-10)
Moonshine: Speech Recognition for Live Transcription and Voice Commands

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Latency-critical speech applications-including live transcription, voice commands, and realtime translation-demand low time-to-first-token (TTFT) and high transcription accuracy, particularly on resource-constrained edge devices. Fullattention Transformer encoders remain a strong accuracy baseline for automatic speech recognition (ASR) because every frame can directly attend to every other frame, which resolves otherwise locally ambiguous acoustics using distant lexical context. However, this global dependency incurs quadratic complexity in sequence length, inducing an inherent "encode-the-wholeutterance" latency profile. For streaming use cases, this causes TTFT to grow linearly with utterance length as the encoder must process the entire prefix before any decoder token can be emitted. To better meet the needs of on-device, streaming ASR use cases we introduce Moonshine v2, an ergodic streaming-encoder ASR model that employs sliding-window self-attention to achieve bounded, low-latency inference while preserving strong local context. Our models achieve state of the art word error rates across standard benchmarks, attaining accuracy on-par with models 6x their size while running significantly faster. These results demonstrate that carefully designed local attention is competitive with the accuracy of full attention at a fraction of the size and latency cost, opening new possibilities for interactive speech interfaces on edge devices.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2602.12241")
get_code_for_paper("2602.12241")
have("2602.12241")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2602.12241.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2602.12241.svg)](https://syntology.ai/paper/2602.12241)

Connect an agent — have() is free.