SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2207.09747 · 2022

Transfer Learning of wav2vec 2.0 for Automatic Lyric Transcription

arXiv · PDF · Open in the Atlas

Code that ran

6 of 7 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; 3 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 1 failure 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/2207.09747.json · badge: /api/ran/2207.09747.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
guxm2021/alt_speechbrain canonical (classifier label)
classifier basis not recorded on the link
6 ran · 1 no recorded run
FunctionStatusWhere it lives
ctc_loss Ran
ran on a synthesized input
guxm2021/alt_speechbrain/speechbrain/nnet/losses.py
code served (permissive licence) · get_code("7860cf2775dccb29")
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 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 torch.nn.functional as F

torch.manual_seed(0)

# Small batch size and dimensions
batch_size = 2
time_steps = 5
num_chars = 4
max_target_len = 3

# log_probs: shape [batch, time, chars]
log_probs = torch.randn(batch_size, time_steps, num_chars, requires_grad=True)
log_probs = F.log_softmax(log_probs, dim=2)

# targets: shape [batch, target_len]
# target_lens will be used to trim this, so we make it longer than needed
targets = torch.tensor([
    [0, 1, 2],
    [1, 2, 0]
], dtype=torch.long)

# input_lens: normalized lengths (will be multiplied by time_steps)
input_lens = torch.tensor([1.0, 0.8], dtype=torch.float)

# target_lens: normalized lengths (will be multiplied by max_target_len)
target_lens = torch.tensor([1.0, 1.0], dtype=torch.float)

# blank_index: the blank symbol index
blank_index = 3

# reduction: one of 'mean', 'sum', 'batch', 'batchmean', 'none'
reduction = 'mean'

ARGS = {
    'log_probs': log_probs,
    'targets': targets,
    'input_lens': input_lens,
    'target_lens': target_lens,
    'blank_index': blank_index,
    'reduction': reduction,
}
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 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_padding_elem Ran
ran, honoured the contract we drafted
guxm2021/alt_speechbrain/speechbrain/nnet/CNN.py
code served (permissive licence) · get_code("8feeecfbeb765eab")
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 int; 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 math

# Construct arguments for get_padding_elem
# The function takes: L_in (int), stride (int), kernel_size (int), dilation (int)

L_in = 64
stride = 2
kernel_size = 5
dilation = 1

ARGS = {
    "L_in": L_in,
    "stride": stride,
    "kernel_size": kernel_size,
    "dilation": dilation,
}
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 int — 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
get_padding_elem_transposed Ran
ran, honoured the contract we drafted
guxm2021/alt_speechbrain/speechbrain/nnet/CNN.py
code served (permissive licence) · get_code("26b6efc761f6e025")
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 torch
import numpy as np

torch.manual_seed(0)

# Based on the function signature, we need to provide integer arguments
# that represent transposed convolution padding computation parameters

ARGS = {
    "L_out": 28,           # output length
    "L_in": 14,            # input length
    "stride": 2,           # stride of transposed convolution
    "kernel_size": 3,      # kernel size
    "dilation": 1,         # dilation factor
    "output_padding": 0,   # output padding
}
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 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
pack_padded_sequence Ran
ran; our contract draft was wrong, not the code
guxm2021/alt_speechbrain/speechbrain/nnet/RNN.py
code served (permissive licence) · get_code("4d856b195967f759")
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 PackedSequence. 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)

# Create inputs tensor: batch_size=2, sequence_length=5, feature_dim=4
inputs = torch.randn(2, 5, 4)

# Create lengths tensor: batch_size=2, values between 0 and 1 (normalized lengths)
# These will be multiplied by inputs.size(1) which is 5
lengths = torch.tensor([0.8, 0.6])

ARGS = {
    "inputs": inputs,
    "lengths": lengths,
}
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 PackedSequence — a PackedSequence 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
pad_packed_sequence Ran
ran, honoured the contract we drafted
guxm2021/alt_speechbrain/speechbrain/nnet/RNN.py
code served (permissive licence) · get_code("7c9d9bad06c577fb")
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, 3, 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
import torch.nn as nn

torch.manual_seed(0)

# Create a PackedSequence object
# PackedSequence requires: data, batch_sizes, sorted_indices, unsorted_indices
# We'll create sequences with batch_size=2, feature_dim=4, variable lengths

# Create raw data: flattened tensor of all timesteps across all sequences
data = torch.randn(5, 4)  # Total 5 timesteps across batch, 4 features each

# Batch sizes: first sequence has 3 timesteps, second has 2
batch_sizes = torch.tensor([2, 2, 1])  # 2 sequences in first timestep, 2 in second, 1 in third

# Create sorted_indices and unsorted_indices
sorted_indices = torch.tensor([0, 1])
unsorted_indices = torch.tensor([0, 1])

# Construct PackedSequence
packed_sequence = torch.nn.utils.rnn.PackedSequence(
    data=data,
    batch_sizes=batch_sizes,
    sorted_indices=sorted_indices,
    unsorted_indices=unsorted_indices
)

ARGS = {
    "inputs": packed_sequence
}
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, 3, 4) — shape (2, 3, 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
parse_arguments Ran
ran on a synthesized input
guxm2021/alt_speechbrain/speechbrain/core.py
code served (permissive licence) · get_code("85608dd83e84cc4e")
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 1, dtype family float; no output shape is stated. 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 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 torch
import argparse

torch.manual_seed(0)

# The function parse_arguments takes one parameter: arg_list
# arg_list should be a list of command-line arguments (strings)
# If None, it reads from sys.argv[1:]

# Construct a valid arg_list with:
# 1. A param_file (required positional argument)
# 2. Various optional arguments that the parser accepts

arg_list = [
    'hyperparams.yaml',  # param_file (required)
    '--debug',
    '--debug_batches', '5',
    '--debug_epochs', '3',
    '--log_config', 'log_config.yaml',
    '--device', 'cpu',
    '--distributed_backend', 'gloo',
    '--find_unused_parameters',
    '--auto_mix_prec',
    '--max_grad_norm', '1.0',
    '--nonfinite_patience', '3',
    '--noprogressbar',
    '--ckpt_interval_minutes', '30.0',
    '--grad_accumulation_factor', '2',
    '--optimizer_step_limit', '1000',
]

ARGS = {
    'arg_list': arg_list
}
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, ndim 1, 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
transducer_loss Not yet run
unverified (harvested, no recorded run)
guxm2021/alt_speechbrain/speechbrain/nnet/losses.py
code served (permissive licence) · get_code("1fab9869e1f0ddf0")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 7 are also present in other papers whose code we hold (7 with at least one paper dated earlier, 0 the same month, 4 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
get_padding_elem 3 other papers
2 dated earlier, 1 later
2005.07143 (2020-05)
ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification
2106.04624 (2021-06)
SpeechBrain: A General-Purpose Speech Toolkit
2303.12659 (2023-03)
Posthoc Interpretation via Quantization
get_padding_elem_transposed 3 other papers
2 dated earlier, 1 later
2005.07143 (2020-05)
ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification
2106.04624 (2021-06)
SpeechBrain: A General-Purpose Speech Toolkit
2303.12659 (2023-03)
Posthoc Interpretation via Quantization
pack_padded_sequence 3 other papers
2 dated earlier, 1 later
2005.07143 (2020-05)
ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification
2106.04624 (2021-06)
SpeechBrain: A General-Purpose Speech Toolkit
2303.12659 (2023-03)
Posthoc Interpretation via Quantization
pad_packed_sequence 3 other papers
2 dated earlier, 1 later
2005.07143 (2020-05)
ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification
2106.04624 (2021-06)
SpeechBrain: A General-Purpose Speech Toolkit
2303.12659 (2023-03)
Posthoc Interpretation via Quantization
ctc_loss 1 other paper
1 dated earlier
2106.04624 (2021-06)
SpeechBrain: A General-Purpose Speech Toolkit
parse_arguments 1 other paper
1 dated earlier
2106.04624 (2021-06)
SpeechBrain: A General-Purpose Speech Toolkit
transducer_loss 1 other paper
1 dated earlier
2106.04624 (2021-06)
SpeechBrain: A General-Purpose Speech Toolkit

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

Automatic speech recognition (ASR) has progressed significantly in recent years due to the emergence of large-scale datasets and the self-supervised learning (SSL) paradigm. However, as its counterpart problem in the singing domain, the development of automatic lyric transcription (ALT) suffers from limited data and degraded intelligibility of sung lyrics. To fill in the performance gap between ALT and ASR, we attempt to exploit the similarities between speech and singing. In this work, we propose a transfer-learning-based ALT solution that takes advantage of these similarities by adapting wav2vec 2.0, an SSL ASR model, to the singing domain. We maximize the effectiveness of transfer learning by exploring the influence of different transfer starting points. We further enhance the performance by extending the original CTC model to a hybrid CTC/attention model. Our method surpasses previous approaches by a large margin on various ALT benchmark datasets. Further experiments show that, with even a tiny proportion of training data, our method still achieves competitive performance.

For agents

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

get_harvested_code_for_paper("2207.09747")
get_code_for_paper("2207.09747")
have("2207.09747")

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

curl https://syntology.ai/api/ran/2207.09747.json

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

[![Syntology run record](https://syntology.ai/api/ran/2207.09747.svg)](https://syntology.ai/paper/2207.09747)

Connect an agent — have() is free.