SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2309.17352 · 2023

Improving Audio Captioning Models with Fine-grained Audio Features, Text Embedding Supervision, and LLM Mix-up Augmentation

arXiv · PDF · Open in the Atlas

Code that ran

10 of 11 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; 10 failures of our instrument; 1 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; 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/2309.17352.json · badge: /api/ran/2309.17352.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
slseanwu/beats-conformer-bart-audio-captioner canonical (classifier label)
classifier basis: census
10 ran · 1 no recorded run
FunctionStatusWhere it lives
collect_audio_text_cosine_sims Ran
ran; our contract draft was wrong, not the code
slseanwu/beats-conformer-bart-audio-captioner/reranking/encoder_rerank_sampling_outputs.py
code served (permissive licence) · get_code("b76e5f7f5d14bd86")
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. 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 torch
import torch.nn as nn
import numpy as np

torch.manual_seed(0)

# Mock instructor model that encodes text to embeddings
class MockInstructorModel:
    def encode(self, inputs, show_progress_bar=False, batch_size=128):
        # inputs is a list of [instruction, caption] pairs
        # Return embeddings with shape (num_inputs, embedding_dim)
        num_inputs = len(inputs)
        embedding_dim = 5
        return np.random.randn(num_inputs, embedding_dim).astype(np.float32)

# Mock audio model that extracts audio embeddings
class MockAudioModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 5)
    
    def get_audio_embed_from_audio(self, encoder_input, attention_mask):
        # encoder_input shape: (batch_size, seq_len)
        # attention_mask shape: (batch_size, seq_len)
        # Return audio embeddings with shape (batch_size, embedding_dim)
        batch_size = encoder_input.shape[0]
        embedding_dim = 5
        # Simple mock: just return random embeddings
        return torch.randn(batch_size, embedding_dim)

# Create model instances
mock_audio_model = MockAudioModel()
mock_instructor_model = MockInstructorModel()

# Create sample data dict with encoder_input and attention_mask
samp_data = {
    'encoder_input': np.random.randint(0, 100, size=(3,)).tolist(),  # 1D array simulating encoded audio
    'attention_mask': np.ones((3,), dtype=np.int64).tolist()  # 1D attention mask
}

# Create a list of generated captions
gen_caption = ['a dog barking', 'a car horn', 'music playing']

ARGS = {
    'model': mock_audio_model,
    'instructor_model': mock_instructor_model,
    'samp_data': samp_data,
    'gen_caption': gen_caption
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn, torch — 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
gelu Ran
ran; our contract draft was wrong, not the code
slseanwu/beats-conformer-bart-audio-captioner/model/modules.py
code served (permissive licence) · get_code("f30ebf4e48b7e2d7")
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 Tensor, ndim 2, dtype family float, shape (2, 4) 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

torch.manual_seed(0)

# Create a small tensor with shape (2, 4)
x = torch.randn(2, 4, dtype=torch.float32)

ARGS = {
    "x": x
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) 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
gelu_accurate Ran
ran; our contract draft was wrong, not the code
slseanwu/beats-conformer-bart-audio-captioner/model/modules.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from BytedanceSpeech/seed-tts-eval, is pointer only (licence: NONE) · get_code("4aa7333ffd7e18c6")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from BytedanceSpeech/seed-tts-eval, and GitHub detected no licence in that repository when it was checked. 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.

generate_caption_for_audio Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
slseanwu/beats-conformer-bart-audio-captioner/inference_sampling.py
code served (permissive licence) · get_code("cc98488959676d1e")
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. 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 torch
import torch.nn as nn
import numpy as np
from unittest.mock import Mock

torch.manual_seed(0)

# Define a minimal PreTrainedModel that satisfies the function's usage
class MinimalAudioModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 8)
    
    def encode_audio(self, encoder_input, attention_mask):
        # encoder_input shape: (batch, seq_len)
        # Returns an object with attention_mask attribute
        output = self.linear(encoder_input)
        result = Mock()
        result.attention_mask = attention_mask
        return result
    
    def generate(self, encoder_outputs, attention_mask, **kwargs):
        # Return a tensor of token sequences (batch, seq_len)
        batch_size = attention_mask.shape[0]
        seq_len = 5
        return torch.randint(0, 100, (batch_size, seq_len))

# Define a minimal PreTrainedTokenizer that satisfies the function's usage
class MinimalTokenizer:
    def decode(self, token_seq, skip_special_tokens=True):
        # Convert token sequence to text
        return "a cat sat on the mat"

# Create model and tokenizer instances
model = MinimalAudioModel()
tokenizer = MinimalTokenizer()

# Create inputs dict with encoder_input, attention_mask, and sample_name
inputs = {
    'encoder_input': np.random.randn(3, 4).astype(np.float32),  # (seq_len, features)
    'attention_mask': np.ones((3,), dtype=np.int64),  # (seq_len,)
    'sample_name': 'test_sample_001'
}

# Create inference config
inference_config = {
    'max_length': 50,
    'num_beams': 1,
    'do_sample': False
}

ARGS = {
    'model': model,
    'tokenizer': tokenizer,
    'inputs': inputs,
    'inference_config': inference_config,
    'strip_punct': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, transformers — 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
mask_along_axis Ran
ran; our contract draft was wrong, not the code
slseanwu/beats-conformer-bart-audio-captioner/model/specaug.py
code served (permissive licence) · get_code("76b744aeab651517")
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 torch

torch.manual_seed(0)

# spec: (Batch, Length, Freq) - a 3D spectrogram tensor
spec = torch.randn(2, 4, 3)  # batch=2, length=4, freq=3

# spec_lengths: (Length) - tensor of lengths for each batch element
spec_lengths = torch.tensor([4, 4])

# mask_width_range: (min, max) for random mask width selection
mask_width_range = (0, 30)

# dim: which dimension to apply mask along (1 or 2)
dim = 1

# num_mask: number of masks to apply per batch element
num_mask = 2

# replace_with_zero: whether to replace masked values with 0 or mean
replace_with_zero = True

ARGS = {
    "spec": spec,
    "spec_lengths": spec_lengths,
    "mask_width_range": mask_width_range,
    "dim": dim,
    "num_mask": num_mask,
    "replace_with_zero": replace_with_zero,
}
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 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
read_inference_csv Ran
ran; our contract draft was wrong, not the code
slseanwu/beats-conformer-bart-audio-captioner/evaluate.py
code served (permissive licence) · get_code("b1a20d3e3e5ca99a")
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, ndim 2, 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 pandas as pd
import tempfile
import os

# Create a temporary CSV file with the required columns
temp_dir = tempfile.mkdtemp()
csv_path = os.path.join(temp_dir, "inference.csv")

# Create a sample DataFrame and save it to CSV
df = pd.DataFrame({
    'file_name': ['image_001.jpg', 'image_002.jpg'],
    'caption_predicted': ['a cat sitting on a mat', 'a dog running in the park']
})
df.to_csv(csv_path, index=False)

ARGS = {
    'csv_path': csv_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 2, dtype family float — a tuple 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
reformat_to_coco Ran
ran; our contract draft was wrong, not the code
slseanwu/beats-conformer-bart-audio-captioner/caption_evaluation_tools/eval_metrics.py
code served (permissive licence) · get_code("9253071bb73fb8b2")
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 torch
import numpy as np
from typing import Dict, List, Union, Tuple, Any

torch.manual_seed(0)

# Define the arguments for reformat_to_coco
predictions = [
    "a dog barking in the distance",
    "birds chirping in a forest"
]

ground_truths = [
    ["a dog barking loudly", "sound of a barking dog", "dog barks"],
    ["chirping birds", "forest birds singing", "birds in nature"]
]

ids = [101, 102]

ARGS = {
    "predictions": predictions,
    "ground_truths": ground_truths,
    "ids": ids
}
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 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
rerank_captions_with_cosine_sim Ran
ran; our contract draft was wrong, not the code
slseanwu/beats-conformer-bart-audio-captioner/reranking/encoder_rerank_sampling_outputs.py
code served (permissive licence) · get_code("76b6ceba370cc6b3")
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. 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

# Set up the arguments for rerank_captions_with_cosine_sim
cosine_sims = np.array([0.85, 0.72, 0.91, 0.68])

samp_idx = 42

samp_name = "audio_sample_001.wav"

samp_true_captions = [
    "A dog barking in the distance",
    "Dogs playing in a park"
]

samp_generated_captions = [
    "A dog barking loudly",
    "Dogs running around",
    "A canine sound in background",
    "Barking sounds from animals"
]

samp_sources = [
    "model_v1",
    "model_v2",
    "model_v1",
    "model_v3"
]

ARGS = {
    "cosine_sims": cosine_sims,
    "samp_idx": samp_idx,
    "samp_name": samp_name,
    "samp_true_captions": samp_true_captions,
    "samp_generated_captions": samp_generated_captions,
    "samp_sources": samp_sources
}
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 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
rerank_captions_with_nll Ran
ran; our contract draft was wrong, not the code
slseanwu/beats-conformer-bart-audio-captioner/reranking/decoder_rerank_sampling_outputs.py
code served (permissive licence) · get_code("bc55a35bc0483eb0")
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. 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 sample data for the function
nlls = np.array([0.5, 0.2, 0.8, 0.3])
samp_idx = 42
samp_name = "audio_sample_001.wav"
samp_true_captions = ["a dog barking", "the dog is barking loudly"]
samp_generated_captions = [
    "a dog barking in the distance",
    "dog barks loudly",
    "the sound of a barking dog",
    "a dog is barking"
]
samp_sources = ["model_a", "model_b", "model_c", "model_d"]

ARGS = {
    "nlls": nlls,
    "samp_idx": samp_idx,
    "samp_name": samp_name,
    "samp_true_captions": samp_true_captions,
    "samp_generated_captions": samp_generated_captions,
    "samp_sources": samp_sources
}
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 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
time_warp Ran
ran; our contract draft was wrong, not the code
slseanwu/beats-conformer-bart-audio-captioner/model/specaug.py
code served (permissive licence) · get_code("460bae18834bc026")
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 Tensor, ndim 3, dtype family float, shape (2, 150, 4) 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

torch.manual_seed(0)

# Construct a small 4D tensor: (Batch, Channel, Time, Freq)
# The function expects (Batch, Time, Freq) but internally adds a Channel dimension
# Shape: (batch=2, time=150, freq=4)
x = torch.randn(2, 150, 4)

ARGS = {
    'x': x,
    'window': 40,
    'mode': 'bicubic',
}
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 Tensor, ndim 3, dtype family float, shape (2, 150, 4) — shape (2, 150, 4) 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_activation_fn Not yet run
unverified (harvested, no recorded run)
slseanwu/beats-conformer-bart-audio-captioner/model/modules.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from BytedanceSpeech/seed-tts-eval, is pointer only (licence: NONE) · get_code("01233f4856456c5f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from BytedanceSpeech/seed-tts-eval, and GitHub detected no licence in that repository when it was checked. 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 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 4 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 3 the same month, 4 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
gelu 11 other papers
5 dated earlier, 1 the same month, 5 later
2204.02152 (2022-04)
UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022
2204.03339 (2022-04)
Boosting Self-Supervised Embeddings for Speech Enhancement
2210.17016 (2022-10)
Wespeaker: A Research and Production oriented Speaker Embedding Learning Toolkit
and 8 more not named here
gelu_accurate 11 other papers
5 dated earlier, 1 the same month, 5 later
2204.02152 (2022-04)
UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022
2204.03339 (2022-04)
Boosting Self-Supervised Embeddings for Speech Enhancement
2210.17016 (2022-10)
Wespeaker: A Research and Production oriented Speaker Embedding Learning Toolkit
and 8 more not named here
get_activation_fn 11 other papers
5 dated earlier, 1 the same month, 5 later
2204.02152 (2022-04)
UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022
2204.03339 (2022-04)
Boosting Self-Supervised Embeddings for Speech Enhancement
2210.17016 (2022-10)
Wespeaker: A Research and Production oriented Speaker Embedding Learning Toolkit
and 8 more not named here
reformat_to_coco 3 other papers
3 later
2312.09818 (2023-12)
SMILE: Multimodal Dataset for Understanding Laughter in Video with Language Models
2403.18572 (2024-03)
ACES: Evaluating Automated Audio Captioning Models on the Semantics of Sounds
2411.18953 (2024-11)
AudioSetCaps: An Enriched Audio-Caption Dataset using Automated Generation Pipeline with Large Audio and Language Models

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

Automated audio captioning (AAC) aims to generate informative descriptions for various sounds from nature and/or human activities. In recent years, AAC has quickly attracted research interest, with state-of-the-art systems now relying on a sequence-to-sequence (seq2seq) backbone powered by strong models such as Transformers. Following the macro-trend of applied machine learning research, in this work, we strive to improve the performance of seq2seq AAC models by extensively leveraging pretrained models and large language models (LLMs). Specifically, we utilize BEATs to extract fine-grained audio features. Then, we employ Instructor LLM to fetch text embeddings of captions, and infuse their language-modality knowledge into BEATs audio features via an auxiliary InfoNCE loss function. Moreover, we propose a novel data augmentation method that uses ChatGPT to produce caption mix-ups (i.e., grammatical and compact combinations of two captions) which, together with the corresponding audio mixtures, increase not only the amount but also the complexity and diversity of training data. During inference, we propose to employ nucleus sampling and a hybrid reranking algorithm, which has not been explored in AAC research. Combining our efforts, our model achieves a new state-of-the-art 32.6 SPIDEr-FL score on the Clotho evaluation split, and wins the 2023 DCASE AAC challenge.

For agents

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

get_harvested_code_for_paper("2309.17352")
get_code_for_paper("2309.17352")
have("2309.17352")

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

curl https://syntology.ai/api/ran/2309.17352.json

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

[![Syntology run record](https://syntology.ai/api/ran/2309.17352.svg)](https://syntology.ai/paper/2309.17352)

Connect an agent — have() is free.