SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2212.09019 · 2022

Fast FullSubNet: Accelerate Full-band and Sub-band Fusion Model for Single-channel Speech Enhancement

arXiv · PDF · Open in the Atlas

Code that ran

9 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; 2 honoured the contract we drafted, 0 violated it and 6 have no contract verdict; 1 failure of our instrument; 2 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/2212.09019.json · badge: /api/ran/2212.09019.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
audio-westlakeu/fullsubnet canonical (classifier label)
classifier basis not recorded on the link
9 ran · 2 no recorded run
FunctionStatusWhere it lives
SI_SDR Ran
ran, honoured the contract we drafted
audio-westlakeu/fullsubnet/audio_zen/metrics.py
code served (permissive licence) · get_code("21bfd01fb48cb3b3")
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 1, dtype family float, shape (2,) 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 reference and estimation signals
# SI_SDR expects numpy arrays with shape [..., T] where T is the time dimension
# Using batch size 2, with time dimension 16000 (1 second at 16kHz)

np.random.seed(0)

# Create reference signal (clean audio)
reference = np.random.randn(2, 16000).astype(np.float32)

# Create estimation signal (processed audio, similar to reference but with some noise)
estimation = reference + 0.1 * np.random.randn(2, 16000).astype(np.float32)

# Sampling rate parameter
sr = 16000

ARGS = {
    "reference": reference,
    "estimation": estimation,
    "sr": sr,
}
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 1, dtype family float, shape (2,) — shape (2,) 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
apply_filterbank Ran
ran on a synthesized input
audio-westlakeu/fullsubnet/audio_zen/acoustics/filtebank.py
code served (permissive licence) · get_code("73a5db3a96cba5c8")
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, 5, 3) 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)

# specgram: [..., F, T]
# Shape: (batch, freq, time) = (2, 4, 3)
specgram = torch.randn(2, 4, 3)

# filterbank: [F, N] where F matches specgram's frequency dimension
# N is the number of filters
# Shape: (freq, num_filters) = (4, 5)
filterbank = torch.randn(4, 5)

ARGS = {
    "specgram": specgram,
    "filterbank": filterbank,
}
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, 5, 3) — shape (2, 5, 3) 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
compress_cIRM Ran
ran, honoured the contract we drafted
audio-westlakeu/fullsubnet/audio_zen/acoustics/mask.py
code served (permissive licence) · get_code("fad2344efce4142b")
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
import numpy as np

torch.manual_seed(0)

# Create test mask as a torch tensor
mask_torch = torch.randn(2, 4)

ARGS = {
    "mask": mask_torch,
    "K": 10,
    "C": 0.1
}
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 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
decompress_cIRM Ran
ran; our contract draft was wrong, not the code
audio-westlakeu/fullsubnet/audio_zen/acoustics/mask.py
code served (permissive licence) · get_code("8dcc65c57ecccc31")
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 cIRM mask tensor with values in the range [-K, K]
# Shape: (2, 4) for batch size 2 and 4 frequency bins
mask = torch.randn(2, 4) * 5  # Scale to roughly [-10, 10] range

ARGS = {
    "mask": mask,
    "K": 10,
    "limit": 9.9
}
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
initialize_module Ran
ran on a synthesized input
audio-westlakeu/fullsubnet/audio_zen/utils.py
code served (permissive licence) · get_code("68cf2d4234677c00")
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 importlib
from typing import Optional

torch.manual_seed(0)

# Create a simple test module and class to load
class SimpleModel(torch.nn.Module):
    def __init__(self, input_dim: int = 10, output_dim: int = 5):
        super().__init__()
        self.linear = torch.nn.Linear(input_dim, output_dim)
    
    def forward(self, x):
        return self.linear(x)

class SimpleFunction:
    def __init__(self, param1: int = 1, param2: str = "test"):
        self.param1 = param1
        self.param2 = param2

# Create a mock module in sys.modules for testing
import sys
import types

mock_module = types.ModuleType('mock_model')
mock_module.SimpleModel = SimpleModel
mock_module.SimpleFunction = SimpleFunction
sys.modules['mock_model'] = mock_module

# Define ARGS for initialize_module function
ARGS = {
    'path': 'mock_model.SimpleModel',
    'args': {'input_dim': 10, 'output_dim': 5},
    'initialize': True
}
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
inverse_filterbank Ran
ran on a synthesized input
audio-westlakeu/fullsubnet/audio_zen/acoustics/filtebank.py
code served (permissive licence) · get_code("2be9f5174fb8a644")
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, 6, 5) 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)

# filtered_specgram: [..., N, T]
# This is a spectrogram that has been filtered through a filterbank
# Shape: [..., N, T] where N is frequency bins and T is time steps
# Let's use shape (2, 4, 5) meaning batch=2, N=4, T=5
filtered_specgram = torch.randn(2, 4, 5)

# filterbank: [F, N]
# This is the inverse filterbank matrix
# F is the original frequency dimension, N is the filtered dimension
# Shape: [F, N] where F=6 (original bins) and N=4 (filtered bins)
filterbank = torch.randn(6, 4)

ARGS = {
    "filtered_specgram": filtered_specgram,
    "filterbank": filterbank,
}
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, 6, 5) — shape (2, 6, 5) 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
istft Ran
ran on a synthesized input
audio-westlakeu/fullsubnet/audio_zen/acoustics/feature.py
code served (permissive licence) · get_code("7c62a70023980973")
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, 10) 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 complex spectrogram tensor [B, F, T]
# B=2 (batch), F=4 (frequency bins), T=3 (time frames)
batch_size = 2
n_fft = 8
n_freqs = n_fft // 2 + 1  # 5 frequency bins
time_frames = 3

# Create complex features
real_part = torch.randn(batch_size, n_freqs, time_frames)
imag_part = torch.randn(batch_size, n_freqs, time_frames)
features_complex = torch.complex(real_part, imag_part)

# Parameters for ISTFT
hop_length = 2
win_length = 8
length = 10  # expected output length

ARGS = {
    'features': features_complex,
    'n_fft': n_fft,
    'hop_length': hop_length,
    'win_length': win_length,
    'length': length,
    'input_type': 'complex'
}
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, 10) — shape (2, 10) 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
mag_phase Ran
ran on a synthesized input
audio-westlakeu/fullsubnet/audio_zen/acoustics/feature.py
code served (permissive licence) · get_code("2e7ab324da47c617")
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 3, 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 torch

torch.manual_seed(0)

# Create a small complex tensor
# Shape: (2, 4) - batch size 2, 4 elements
complex_tensor = torch.randn(2, 4, dtype=torch.complex64)

ARGS = {
    "complex_tensor": complex_tensor
}
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 3, 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
stft Ran
ran on a synthesized input
audio-westlakeu/fullsubnet/audio_zen/acoustics/feature.py
code served (permissive licence) · get_code("8fe3a6260008040f")
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 4, 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 torch

torch.manual_seed(0)

# Single-channel audio: [B, T]
y_2d = torch.randn(2, 8192)

# Multi-channel audio: [B, C, T]
y_3d = torch.randn(2, 4, 8192)

# STFT parameters
n_fft = 512
hop_length = 256
win_length = 512

# Test with 2D input
ARGS = {
    "y": y_2d,
    "n_fft": n_fft,
    "hop_length": hop_length,
    "win_length": win_length,
}
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 4, 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
load_checkpoint Not yet run
unverified (harvested, no recorded run)
audio-westlakeu/fullsubnet/audio_zen/utils.py
code served (permissive licence) · get_code("21ba83a7fbb72202")
How it ran

no invocation recorded.

print_tensor_info Not yet run
unverified (harvested, no recorded run)
audio-westlakeu/fullsubnet/audio_zen/utils.py
code served (permissive licence) · get_code("3099b5d08820498b")
How it ran

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, 8 are also present in other papers whose code we hold (8 with at least one paper dated earlier, 0 the same month, 1 later) and 3 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
SI_SDR 2 other papers
1 dated earlier, 1 later
2010.15508 (2020-10)
FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement
2305.05599 (2023-05)
Inter-SubNet: Speech Enhancement with Subband Interaction
apply_filterbank 1 other paper
1 dated earlier
2010.15508 (2020-10)
FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement
compress_cIRM 1 other paper
1 dated earlier
2010.15508 (2020-10)
FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement
decompress_cIRM 1 other paper
1 dated earlier
2010.15508 (2020-10)
FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement
inverse_filterbank 1 other paper
1 dated earlier
2010.15508 (2020-10)
FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement
load_checkpoint 1 other paper
1 dated earlier
2010.15508 (2020-10)
FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement
mag_phase 1 other paper
1 dated earlier
2010.15508 (2020-10)
FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement
print_tensor_info 1 other paper
1 dated earlier
2010.15508 (2020-10)
FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement

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

FullSubNet is our recently proposed real-time single-channel speech enhancement network that achieves outstanding performance on the Deep Noise Suppression (DNS) Challenge dataset. A number of variants of FullSubNet have been proposed, but they all focus on the structure design towards better performance and are rarely concerned with computational efficiency. For many speech enhancement applications, a key feature is that system runs on a real-time, latency-sensitive, battery-powered platform, which strictly limits the algorithm latency and computational complexity. In this work, we propose a new architecture named Fast FullSubNet dedicated to accelerating the computation of FullSubNet. Specifically, Fast FullSubNet processes sub-band speech spectra in the mel-frequency domain by using cascaded linear-to-mel full-band, sub-band, and mel-to-linear full-band models such that frequencies involved in the sub-band computation are vastly reduced. After that, a down-sampling operation is proposed for the sub-band input sequence to further reduce the computational complexity along the time axis. Experimental results show that, compared to FullSubNet, Fast FullSubNet has only 13\% computational complexity and 16\% processing time, and achieves comparable or even better performance. Code and audio samples are available at https://github.com/Audio-WestlakeU/FullSubNet.

For agents

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

get_harvested_code_for_paper("2212.09019")
get_code_for_paper("2212.09019")
have("2212.09019")

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

curl https://syntology.ai/api/ran/2212.09019.json

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

[![Syntology run record](https://syntology.ai/api/ran/2212.09019.svg)](https://syntology.ai/paper/2212.09019)

Connect an agent — have() is free.