SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.11933 · 2026

Cross-Modal Robustness Transfer (CMRT): Training Robust Speech Translation Models Using Adversarial Text

Gerasimos Spanakis, Yusuf Can, Abderrahmane Issam, Semerci Scholtes

arXiv · PDF

Code that ran

5 of 6 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; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 5 failures of our instrument; 1 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/2602.11933.json · badge: /api/ran/2602.11933.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
NVIDIA/NeMo canonical (classifier label)
classifier basis not recorded on the link
5 ran · 1 no recorded run
FunctionStatusWhere it lives
analyze_imports Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo/nemo_dependencies.py
code served (permissive licence) · get_code("b440f7f9d47e9715")
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 set. 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 ast
import tempfile
import os

# Create a temporary Python file with NeMo imports to analyze
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test_nemo_imports.py")

# Write a test file with various NeMo imports
test_code = """
from nemo.collections.asr import models
from nemo.collections.nlp.modules import encoders
from nemo.collections.tts import models as tts_models
from nemo.core import types
from nemo.utils import logging
from nemo.export import export_to_onnx
from nemo.deploy import inference
from nemo.lightning import trainer
from nemo.automodel import load_model
import numpy as np
"""

with open(temp_file, 'w', encoding='utf-8') as f:
    f.write(test_code)

ARGS = {
    'file_path': temp_file
}
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 set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set 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
detect_prefix Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo/nemo/utils/model_utils.py
code served (permissive licence) · get_code("133df64d93db327e")
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.

import fnmatch

ARGS = {
    'names': [
        'model_config.yaml',
        'other_file.txt',
        'another_file.bin'
    ]
}
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
find_python_files Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo/nemo_dependencies.py
code served (permissive licence) · get_code("6c5bcba1a5bb7625")
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 list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 os
import tempfile
import shutil

# Create a temporary directory structure for testing
temp_dir = tempfile.mkdtemp()

# Create the directory structure with some Python files
dirs_to_create = ['nemo', 'scripts', 'examples', 'tests']
for dir_name in dirs_to_create:
    dir_path = os.path.join(temp_dir, dir_name)
    os.makedirs(dir_path, exist_ok=True)
    
    # Create some dummy Python files
    with open(os.path.join(dir_path, 'test_file.py'), 'w') as f:
        f.write('# dummy python file\n')
    
    # Create a subdirectory with more Python files
    subdir = os.path.join(dir_path, 'subdir')
    os.makedirs(subdir, exist_ok=True)
    with open(os.path.join(subdir, 'another_file.py'), 'w') as f:
        f.write('# another dummy python file\n')

# Also create a non-relevant directory that should be ignored
other_dir = os.path.join(temp_dir, 'other')
os.makedirs(other_dir, exist_ok=True)
with open(os.path.join(other_dir, 'ignored.py'), 'w') as f:
    f.write('# this should be ignored\n')

ARGS = {
    'directory': temp_dir
}
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 list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
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
find_top_level_packages Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo/nemo_dependencies.py
code served (permissive licence) · get_code("167ef881cd9f9133")
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 list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 os
import tempfile
import shutil

# Create a temporary directory structure to simulate nemo package layout
temp_dir = tempfile.mkdtemp()
nemo_root = temp_dir

# Create the nemo directory with some subdirectories
nemo_dir = os.path.join(nemo_root, 'nemo')
os.makedirs(nemo_dir, exist_ok=True)

# Create some top-level packages (directories)
packages_to_create = ['core', 'models', 'utils', 'collections']
for package in packages_to_create:
    os.makedirs(os.path.join(nemo_dir, package), exist_ok=True)

# Create __pycache__ to test that it's excluded
os.makedirs(os.path.join(nemo_dir, '__pycache__'), exist_ok=True)

# Create a file (not a directory) to test that it's excluded
with open(os.path.join(nemo_dir, 'README.txt'), 'w') as f:
    f.write('test')

ARGS = {
    'nemo_root': nemo_root
}
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 list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
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
unwrap_model Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo/nemo/utils/model_utils.py
code served (permissive licence) · get_code("d0b9c431aaf4be9f")
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 SimpleModel. 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 torch
import torch.nn as nn

torch.manual_seed(0)

# Create a simple base model
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Create wrapper classes to test unwrapping
class Float16Module(nn.Module):
    def __init__(self, module):
        super().__init__()
        self.module = module
    
    def forward(self, x):
        return self.module(x)

class DoubleWrappedModule(nn.Module):
    def __init__(self, module):
        super().__init__()
        self.module = module
    
    def forward(self, x):
        return self.module(x)

# Test case 1: Single wrapped model
base_model = SimpleModel()
wrapped_model = Float16Module(base_model)
double_wrapped_model = DoubleWrappedModule(wrapped_model)

# Test case 2: List of models
model_list = [double_wrapped_model, wrapped_model]

# Test case 3: Custom wrapper type
class CustomWrapper(nn.Module):
    def __init__(self, module):
        super().__init__()
        self.module = module
    
    def forward(self, x):
        return self.module(x)

custom_wrapped = CustomWrapper(SimpleModel())

ARGS = {
    "model": double_wrapped_model,
    "module_instances": (Float16Module, DoubleWrappedModule)
}
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 SimpleModel — a SimpleModel output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a SimpleModel 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
safe_instantiate Not yet run
unverified (harvested, no recorded run)
NVIDIA/NeMo/nemo/core/classes/common.py
code served (permissive licence) · get_code("9f06792a2d142360")
How it ran

no invocation recorded.

Also present in other papers' code

Of 6 samples lifted from this paper's linked repositories, 6 carry a function identity and 0 do not. Of the 6 distinct functions with an identity, 6 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 0 the same month, 6 later) and 0 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
analyze_imports 3 other papers
1 dated earlier, 2 later
2601.13240 (2026-01)
KOCO-BENCH: Can Large Language Models Leverage Domain Knowledge in Software Development?
2606.06047 (2026-06)
Automatic Labelling of Speech Translation Errors
2608.13717 (2026-08)
StreamHear: Domain-Adapted Pseudo-Labeling for Semi-Supervised Streaming Speech Recognition
detect_prefix 3 other papers
1 dated earlier, 2 later
2025.naacl-long.298 (2025 (year only))
VoiceTextBlender: Augmenting Large Language Models with Speech Capabilities via Single-Stage Joint Speech-Text Supervised Fine-Tuning
2606.06047 (2026-06)
Automatic Labelling of Speech Translation Errors
2608.13717 (2026-08)
StreamHear: Domain-Adapted Pseudo-Labeling for Semi-Supervised Streaming Speech Recognition
find_python_files 3 other papers
1 dated earlier, 2 later
2601.13240 (2026-01)
KOCO-BENCH: Can Large Language Models Leverage Domain Knowledge in Software Development?
2606.06047 (2026-06)
Automatic Labelling of Speech Translation Errors
2608.13717 (2026-08)
StreamHear: Domain-Adapted Pseudo-Labeling for Semi-Supervised Streaming Speech Recognition
find_top_level_packages 3 other papers
1 dated earlier, 2 later
2601.13240 (2026-01)
KOCO-BENCH: Can Large Language Models Leverage Domain Knowledge in Software Development?
2606.06047 (2026-06)
Automatic Labelling of Speech Translation Errors
2608.13717 (2026-08)
StreamHear: Domain-Adapted Pseudo-Labeling for Semi-Supervised Streaming Speech Recognition
safe_instantiate 2 other papers
2 later
2606.06047 (2026-06)
Automatic Labelling of Speech Translation Errors
2608.13717 (2026-08)
StreamHear: Domain-Adapted Pseudo-Labeling for Semi-Supervised Streaming Speech Recognition
unwrap_model 2 other papers
2 later
2606.06047 (2026-06)
Automatic Labelling of Speech Translation Errors
2608.13717 (2026-08)
StreamHear: Domain-Adapted Pseudo-Labeling for Semi-Supervised Streaming Speech Recognition

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

End-to-End Speech Translation (E2E-ST) has seen significant advancements, yet current models are primarily benchmarked on curated, "clean" datasets. This overlooks critical realworld challenges, such as morphological robustness to inflectional variations common in non-native or dialectal speech. In this work, we adapt a text-based adversarial attack targeting inflectional morphology to the speech domain and demonstrate that state-of-the-art E2E-ST models are highly vulnerable it. While adversarial training effectively mitigates such risks in text-based tasks, generating high-quality adversarial speech data remains computationally expensive and technically challenging. To address this, we propose Cross-Modal Robustness Transfer (CMRT), a framework that transfers adversarial robustness from the text modality to the speech modality. Our method eliminates the requirement for adversarial speech data during training. Extensive experiments across four language pairs demonstrate that CMRT improves adversarial robustness by an average of more than 3 BLEU points, establishing a new baseline for robust E2E-ST without the overhead of generating adversarial speech.

For agents

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

get_harvested_code_for_paper("2602.11933")
get_code_for_paper("2602.11933")
have("2602.11933")

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

curl https://syntology.ai/api/ran/2602.11933.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.11933.svg)](https://syntology.ai/paper/2602.11933)

Connect an agent — have() is free.