SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.07143 · 2026

Massive Sound Embedding Benchmark (MSEB)

Georg Heigold, Michael Riley, Ji Ma, Shankar Kumar, Ehsan Variani, Tom Bagby, Cyril Allauzen

arXiv · PDF · Open in the Atlas

Code that ran

6 of 8 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; 2 failures of our instrument; 2 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/2602.07143.json · badge: /api/ran/2602.07143.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
google-research/mseb canonical (classifier label)
classifier basis: census
6 ran · 2 no recorded run
FunctionStatusWhere it lives
compute_average_precision Ran
ran, honoured the contract we drafted
google-research/mseb/mseb/metrics.py
code served (permissive licence) · get_code("2039ed771f5c6e3c")
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.

# Setup code for compute_average_precision

ARGS = {
    "reference": ["doc1", "doc2", "doc3"],
    "predicted_neighbors": ["doc2", "doc4", "doc1", "doc5", "doc3", "doc6"]
}
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
compute_exact_match Ran
ran, honoured the contract we drafted
google-research/mseb/mseb/metrics.py
code served (permissive licence) · get_code("db44f20e41014fcf")
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
from collections.abc import Sequence

torch.manual_seed(0)

# Test case 1: reference as a single string, predicted_neighbors with a match
reference_1 = "apple"
predicted_neighbors_1 = ["apple", "banana", "cherry"]

# Test case 2: reference as a sequence of strings, predicted_neighbors with a match
reference_2 = ["apple", "apricot", "avocado"]
predicted_neighbors_2 = ["apricot", "orange", "grape"]

# Test case 3: reference as a string, predicted_neighbors with no match
reference_3 = "dog"
predicted_neighbors_3 = ["cat", "bird", "fish"]

# Test case 4: empty predicted_neighbors
reference_4 = "test"
predicted_neighbors_4 = []

# Use the first test case for ARGS
ARGS = {
    "reference": reference_1,
    "predicted_neighbors": predicted_neighbors_1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
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
compute_reciprocal_rank Ran
ran, honoured the contract we drafted
google-research/mseb/mseb/metrics.py
code served (permissive licence) · get_code("9f778f299692103e")
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
from collections.abc import Sequence

torch.manual_seed(0)

# Construct valid arguments for compute_reciprocal_rank
ARGS = {
    "reference": "apple",
    "predicted_neighbors": ["banana", "cherry", "apple", "date"]
}
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_base_path Ran
ran; our contract draft was wrong, not the code
google-research/mseb/mseb/dataset.py
code served (permissive licence) · get_code("02b1383aeb11c226")
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.

from absl import flags

# Initialize flags to avoid issues with flag definitions
flags.FLAGS.mark_as_parsed()

# Define the flag if not already defined
try:
    _DATASET_BASEPATH = flags.DEFINE_string('dataset_basepath', None, 'Path to the MSEB dataset cache.')
except flags.DuplicateFlagError:
    _DATASET_BASEPATH = flags.FLAGS['dataset_basepath']

# Test case 1: basepath provided as argument
ARGS = {
    'basepath': '/path/to/dataset'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
absl — 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_num_shards Ran
ran, honoured the contract we drafted
google-research/mseb/mseb/runner.py
code served (permissive licence) · get_code("c2fd7527699739bd")
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 int, ndim 0, dtype family int, 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 math

def get_num_shards(num_examples: int, examples_per_shard: int=25000) -> int:
    return math.ceil(num_examples / examples_per_shard)

ARGS = {
    "num_examples": 100000,
    "examples_per_shard": 25000
}
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 int, ndim 0, dtype family int, 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
parse_resource_hints Ran
ran; our contract draft was wrong, not the code
google-research/mseb/mseb/runner.py
code served (permissive licence) · get_code("52f384c4d6baf04b")
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 dict. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 json
from collections.abc import Mapping

cpu_resource_hints = dict()

ARGS = {
    "resource_hints": '{"gpu": 1, "memory": "4GB"}',
    "accelerator": "cuda"
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
list_hf_files Not yet run
unverified (harvested, no recorded run)
google-research/mseb/mseb/utils.py
code served (permissive licence) · get_code("0b378aacda0c34a4")
How it ran

no invocation recorded.

read_hf_parquet Not yet run
unverified (harvested, no recorded run)
google-research/mseb/mseb/utils.py
code served (permissive licence) · get_code("472187cb76b518be")
How it ran

no invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 8 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.

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

Audio is a critical component of multimodal perception, and any truly intelligent system must demonstrate a wide range of auditory capabilities. These capabilities include transcription, classification, retrieval, reasoning, segmentation, clustering, reranking, and reconstruction. Fundamentally, each task involves transforming a raw audio signal into a meaningful 'embedding'-be it a single vector, a sequence of continuous or discrete representations, or another structured form-which then serves as the basis for generating the task's final response. To accelerate progress towards robust machine auditory intelligence, we present the Massive Sound Embedding Benchmark (MSEB): an extensible framework designed to evaluate the auditory components of any multimodal system. In its first release, MSEB offers a comprehensive suite of eight core tasks, with more planned for the future, supported by diverse datasets, including the new, large-scale Simple Voice Questions (SVQ) dataset. Our initial experiments establish clear performance headrooms, highlighting the significant opportunity to improve real-world multimodal experiences where audio is a core signal. We encourage the research community to use MSEB to assess their algorithms and contribute to its growth. The library is publicly hosted at https://github.com/google-research/mseb. * "question": The question being asked (string). * "title": The title of the document the context is from (string). * "context": A text passage which can either be a wiki page or a wiki paragraph that may or may not contain the answer. **Output:** You will produce a single JSON object as a plain text string (no markup). The structure depends on answerability: If the question IS answerable: * "rationale": (string) A concise explanation of why the provided answer is correct. Be specific, referencing sentences or phrases. * "answer": (string) The answer to the question, copied **exactly** from the title or context. Do not paraphrase or summarize. Prefer concise answers (shortest possible while complete). If the question IS NOT answerable: * "No Answer": (string) A clear and concise explanation of why the question cannot be answered. Specify what information is missing. **Important Considerations:** * Code change: The question may be in a language differ from context and title. In that case, answer the question with the same language as context and title.

For agents

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

get_harvested_code_for_paper("2602.07143")
get_code_for_paper("2602.07143")
have("2602.07143")

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

curl https://syntology.ai/api/ran/2602.07143.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.07143.svg)](https://syntology.ai/paper/2602.07143)

Connect an agent — have() is free.