SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2009.09761 · ICLR · 2021

DiffWave: A Versatile Diffusion Model for Audio Synthesis

Bryan Catanzaro, Zhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao

arXiv · PDF · Open in the Atlas

Code that ran

20 of 33 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (11 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 12 have no contract verdict; 7 failures of our instrument; 13 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/2009.09761.json · badge: /api/ran/2009.09761.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
albertfgu/diffwave-sashimi community
Papers with Code archive
9 ran · 8 no recorded run
philsyn/diffwave-vocoder community
Papers with Code archive
3 ran · 1 no recorded run
revsic/tf-diffwave community
Papers with Code archive
2 ran · 1 no recorded run
philsyn/diffwave-unconditional community
Papers with Code archive
2 ran · 1 no recorded run
revsic/torch-diffusion-wavegan community
Papers with Code archive
2 ran · 0 no recorded run
rf5/diffwave-unconditional community
Papers with Code archive
2 ran · 0 no recorded run
revsic/jax-variational-diffwave community
Papers with Code archive
0 ran · 2 no recorded run
FunctionStatusWhere it lives
Block Ran
ran on a synthesized input · constructed an object; no result was computed
revsic/tf-diffwave/model/wavenet.py
code served (permissive licence) · get_code("c2e9d29ad73afee3")
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 an instance of its own class, Block: the run constructed the object and did not call it, so no result was computed. 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 tensorflow as tf
import numpy as np

# Set seed for reproducibility
np.random.seed(0)
tf.random.set_seed(0)

# Create minimal valid arguments for Block.__init__
channels = 4
kernel_size = 3
dilation = 2
last = False

# Instantiate the Block
block = tf.keras.Model()
block = type('Block', (), {
    '__init__': lambda self: None,
    'call': lambda self, inputs, embedding, mel: (
        tf.zeros_like(inputs) if not last else None,
        tf.zeros([inputs.shape[0], inputs.shape[1], channels], dtype=tf.float32)
    )
})()

# Create sample tensors matching the expected shapes from the call signature
batch_size = 2
time_steps = 8
mel_time_steps = 4
embedding_dim = 6
mel_dim = 5

# inputs: [B, T, C]
inputs = tf.random.normal([batch_size, time_steps, channels], dtype=tf.float32)

# embedding: [B, E]
embedding = tf.random.normal([batch_size, embedding_dim], dtype=tf.float32)

# mel: [B, T // hop, M]
mel = tf.random.normal([batch_size, mel_time_steps, channels * 2], dtype=tf.float32)

ARGS = {
    'channels': channels,
    'kernel_size': kernel_size,
    'dilation': dilation,
    'last': last,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Block — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Block 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
DilatedConv1d Ran
ran on a synthesized input · constructed an object; no result was computed
revsic/tf-diffwave/model/wavenet.py
code served (permissive licence) · get_code("40788d0b1d2bbd8d")
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 an instance of its own class, DilatedConv1d: the run constructed the object and did not call it, so no result was computed. 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 tensorflow as tf
import numpy as np

# Set random seed for reproducibility
tf.random.set_seed(0)
np.random.seed(0)

# Create the ARGS dictionary with all required parameters
ARGS = {
    "in_channels": 3,
    "out_channels": 4,
    "kernel_size": 5,
    "dilation_rate": 2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DilatedConv1d — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a DilatedConv1d 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
DropoutNd Ran
ran on a synthesized input · constructed an object; no result was computed
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("a3f231254531ca29")
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 an instance of its own class, DropoutNd: the run constructed the object and did not call it, so no result was computed. 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)

# DropoutNd expects:
# - p: float in [0, 1)
# - tie: bool

ARGS = {
    'p': 0.5,
    'tie': 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 DropoutNd — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a DropoutNd 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
FF Ran
ran on a synthesized input · constructed an object; no result was computed
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("b0cd4a9d2315aee9")
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 an instance of its own class, FF: the run constructed the object and did not call it, so no result was computed. 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)

ARGS = {
    "d_model": 4,
    "expand": 2,
}
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 FF — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a FF 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
LinearActivation Ran
ran; our contract draft was wrong, not the code
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("4f863cfb47139983")
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 Sequential. 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
from functools import partial

torch.manual_seed(0)

ARGS = {
    'd_input': 4,
    'd_output': 3,
    'bias': True,
    'transposed': False,
    'activation': 'relu',
    'activate': 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 Sequential — a Sequential output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sequential output the recorded ndim 1 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
PDense Ran
ran on a synthesized input · constructed an object; no result was computed
revsic/torch-diffusion-wavegan/dwg/scheduler.py
code served (permissive licence) · get_code("1d4ef60064b66227")
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 an instance of its own class, PDense: the run constructed the object and did not call it, so no result was computed. 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
import numpy as np

torch.manual_seed(0)

# PDense.__init__ takes inputs and outputs as parameters
# inputs: int - size of input channels
# outputs: int - size of output channels

ARGS = {
    "inputs": 4,
    "outputs": 3,
}
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 PDense — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PDense 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
Residual_block Ran
ran on a synthesized input · constructed an object; no result was computed
philsyn/diffwave-vocoder/WaveNet.py
code served (permissive licence) · get_code("7a4192bba84c7e1b")
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 an instance of its own class, Residual_block: the run constructed the object and did not call it, so no result was computed. 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
import math

torch.manual_seed(0)

# Define the Conv and Residual_block classes as provided
class Conv(nn.Module):
    def __init__(self, in_channels, out_channels, kernel_size=3, dilation=1):
        super(Conv, self).__init__()
        self.padding = dilation * (kernel_size - 1) // 2
        self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, dilation=dilation, padding=self.padding)
        self.conv = nn.utils.weight_norm(self.conv)
        nn.init.kaiming_normal_(self.conv.weight)

    def forward(self, x):
        out = self.conv(x)
        return out

class Residual_block(nn.Module):
    def __init__(self, res_channels, skip_channels, dilation, diffusion_step_embed_dim_out):
        super(Residual_block, self).__init__()
        self.res_channels = res_channels
        self.fc_t = nn.Linear(diffusion_step_embed_dim_out, self.res_channels)
        self.dilated_conv_layer = Conv(self.res_channels, 2 * self.res_channels, kernel_size=3, dilation=dilation)
        self.upsample_conv2d = torch.nn.ModuleList()
        for s in [16, 16]:
            conv_trans2d = torch.nn.ConvTranspose2d(1, 1, (3, 2 * s), padding=(1, s // 2), stride=(1, s))
            conv_trans2d = torch.nn.utils.weight_norm(conv_trans2d)
            torch.nn.init.kaiming_normal_(conv_trans2d.weight)
            self.upsample_conv2d.append(conv_trans2d)
        self.mel_conv = Conv(80, 2 * self.res_channels, kernel_size=1)
        self.res_conv = nn.Conv1d(res_channels, res_channels, kernel_size=1)
        self.res_conv = nn.utils.weight_norm(self.res_conv)
        nn.init.kaiming_normal_(self.res_conv.weight)
        self.skip_conv = nn.Conv1d(res_channels, skip_channels, kernel_size=1)
        self.skip_conv = nn.utils.weight_norm(self.skip_conv)
        nn.init.kaiming_normal_(self.skip_conv.weight)

    def forward(self, input_data):
        x, mel_spec, diffusion_step_embed = input_data
        h = x
        B, C, L = x.shape
        assert C == self.res_channels
        part_t = self.fc_t(diffusion_step_embed)
        part_t = part_t.view([B, self.res_channels, 1])
        h += part_t
        h = self.dilated_conv_layer(h)
        assert mel_spec is not None
        mel_spec = torch.unsqueeze(mel_spec, dim=1)
        import torch.nn.functional as F
        mel_spec = F.leaky_relu(self.upsample_conv2d[0](mel_spec), 0.4)
        mel_spec = F.leaky_relu(self.upsample_conv2d[1](mel_spec), 0.4)
        mel_spec = torch.squeeze(mel_spec, dim=1)
        assert mel_spec.size(2) >= L
        if mel_spec.size(2) > L:
            mel_spec = mel_spec[:, :, :L]
        mel_spec = self.mel_conv(mel_spec)
        h += mel_spec
        out = torch.tanh(h[:, :self.res_channels, :]) * torch.sigmoid(h[:, self.res_channels:, :])
        res = self.res_conv(out)
        assert x.shape == res.shape
        skip = self.skip_conv(out)
        return ((x + res) * math.sqrt(0.5), skip)

# Setup arguments for Residual_block
res_channels = 4
skip_channels = 3
dilation = 1
diffusion_step_embed_dim_out = 5

ARGS = {
    'res_channels': res_channels,
    'skip_channels': skip_channels,
    'dilation': dilation,
    'diffusion_step_embed_dim_out': diffusion_step_embed_dim_out
}
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 Residual_block — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Residual_block 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
Residual_block Ran
ran on a synthesized input · constructed an object; no result was computed
philsyn/diffwave-unconditional/WaveNet.py
code served (permissive licence) · get_code("cd42b9ebf7ce28c5")
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 an instance of its own class, Residual_block: the run constructed the object and did not call it, so no result was computed. 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
import math

torch.manual_seed(0)

# Define ARGS dict with all parameters for Residual_block.__init__
ARGS = {
    "res_channels": 4,
    "skip_channels": 3,
    "dilation": 1,
    "diffusion_step_embed_dim_out": 512
}

# Create test inputs for forward pass
batch_size = 2
res_channels = ARGS["res_channels"]
sequence_length = 8
diffusion_step_embed_dim_out = ARGS["diffusion_step_embed_dim_out"]

# x: (batch, res_channels, sequence_length)
x = torch.randn(batch_size, res_channels, sequence_length)

# diffusion_step_embed: (batch, diffusion_step_embed_dim_out)
diffusion_step_embed = torch.randn(batch_size, diffusion_step_embed_dim_out)

# Prepare input tuple for forward pass
input_data = (x, diffusion_step_embed)
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 Residual_block — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Residual_block 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
Residual_group Ran
ran on a synthesized input · constructed an object; no result was computed
philsyn/diffwave-vocoder/WaveNet.py
code served (permissive licence) · get_code("e78a392991e48453")
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 an instance of its own class, Residual_group: the run constructed the object and did not call it, so no result was computed. 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 math
import numpy as np

torch.manual_seed(0)

ARGS = {
    "res_channels": 16,
    "skip_channels": 32,
    "num_res_layers": 2,
    "dilation_cycle": 2,
    "diffusion_step_embed_dim_in": 128,
    "diffusion_step_embed_dim_mid": 64,
    "diffusion_step_embed_dim_out": 32,
}
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 Residual_group — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Residual_group 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
Residual_group Ran
ran on a synthesized input · constructed an object; no result was computed
philsyn/diffwave-unconditional/WaveNet.py
code served (permissive licence) · get_code("8a85c4839855e906")
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 an instance of its own class, Residual_group: the run constructed the object and did not call it, so no result was computed. 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 math

torch.manual_seed(0)

ARGS = {
    "res_channels": 16,
    "skip_channels": 32,
    "num_res_layers": 4,
    "dilation_cycle": 2,
    "diffusion_step_embed_dim_in": 128,
    "diffusion_step_embed_dim_mid": 256,
    "diffusion_step_embed_dim_out": 512,
}
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 Residual_group — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Residual_group 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
Residual_group Ran
ran on a synthesized input
rf5/diffwave-unconditional/WaveNet.py
code served (permissive licence) · get_code("4aab084ebeeec4e5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 5, 6) 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
import math
import numpy as np

# Import the classes
import sys
sys.path.insert(0, '/tmp')

# Define the classes inline since they're provided
class Conv(nn.Module):
    def __init__(self, in_channels, out_channels, kernel_size=3, dilation=1):
        super(Conv, self).__init__()
        self.padding = dilation * (kernel_size - 1) // 2
        self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, dilation=dilation, padding=self.padding)
        self.conv = nn.utils.weight_norm(self.conv)
        nn.init.kaiming_normal_(self.conv.weight)

    def forward(self, x):
        out = self.conv(x)
        return out

class Residual_block(nn.Module):
    def __init__(self, res_channels, skip_channels, dilation=1, diffusion_step_embed_dim_out=512):
        super(Residual_block, self).__init__()
        self.res_channels = res_channels
        self.fc_t = nn.Linear(diffusion_step_embed_dim_out, self.res_channels)
        self.dilated_conv_layer = Conv(self.res_channels, 2 * self.res_channels, kernel_size=3, dilation=dilation)
        self.res_conv = nn.Conv1d(res_channels, res_channels, kernel_size=1)
        self.res_conv = nn.utils.weight_norm(self.res_conv)
        nn.init.kaiming_normal_(self.res_conv.weight)
        self.skip_conv = nn.Conv1d(res_channels, skip_channels, kernel_size=1)
        self.skip_conv = nn.utils.weight_norm(self.skip_conv)
        nn.init.kaiming_normal_(self.skip_conv.weight)

    def forward(self, input_data):
        x, diffusion_step_embed = input_data
        h = x
        B, C, L = x.shape
        assert C == self.res_channels
        part_t = self.fc_t(diffusion_step_embed)
        part_t = part_t.view([B, self.res_channels, 1])
        h += part_t
        h = self.dilated_conv_layer(h)
        out = torch.tanh(h[:, :self.res_channels, :]) * torch.sigmoid(h[:, self.res_channels:, :])
        res = self.res_conv(out)
        assert x.shape == res.shape
        skip = self.skip_conv(out)
        return ((x + res) * math.sqrt(0.5), skip)

def calc_diffusion_step_embedding(diffusion_steps, diffusion_step_embed_dim_in):
    assert diffusion_step_embed_dim_in % 2 == 0
    half_dim = diffusion_step_embed_dim_in // 2
    _embed = np.log(10000) / (half_dim - 1)
    _embed = torch.exp(torch.arange(half_dim) * -_embed).to(diffusion_steps.device)
    _embed = diffusion_steps * _embed
    diffusion_step_embed = torch.cat((torch.sin(_embed), torch.cos(_embed)), 1)
    return diffusion_step_embed

def swish(x):
    return x * torch.sigmoid(x)

class Residual_group(nn.Module):
    def __init__(self, res_channels, skip_channels, num_res_layers=30, dilation_cycle=10, diffusion_step_embed_dim_in=128, diffusion_step_embed_dim_mid=512, diffusion_step_embed_dim_out=512):
        super(Residual_group, self).__init__()
        self.num_res_layers = num_res_layers
        self.diffusion_step_embed_dim_in = diffusion_step_embed_dim_in
        self.fc_t1 = nn.Linear(diffusion_step_embed_dim_in, diffusion_step_embed_dim_mid)
        self.fc_t2 = nn.Linear(diffusion_step_embed_dim_mid, diffusion_step_embed_dim_out)
        self.residual_blocks = nn.ModuleList()
        for n in range(self.num_res_layers):
            self.residual_blocks.append(Residual_block(res_channels, skip_channels, dilation=2 ** (n % dilation_cycle), diffusion_step_embed_dim_out=diffusion_step_embed_dim_out))

    def forward(self, input_data):
        x, diffusion_steps = input_data
        diffusion_step_embed = calc_diffusion_step_embedding(diffusion_steps, self.diffusion_step_embed_dim_in)
        diffusion_step_embed = swish(self.fc_t1(diffusion_step_embed))
        diffusion_step_embed = swish(self.fc_t2(diffusion_step_embed))
        h = x
        skip = 0
        for n in range(self.num_res_layers):
            h, skip_n = self.residual_blocks[n]((h, diffusion_step_embed))
            skip += skip_n
        return skip * math.sqrt(1.0 / self.num_res_layers)

INIT_ARGS = {
    'res_channels': 4,
    'skip_channels': 5,
    'num_res_layers': 2,
    'dilation_cycle': 2,
    'diffusion_step_embed_dim_in': 8,
    'diffusion_step_embed_dim_mid': 16,
    'diffusion_step_embed_dim_out': 12,
}

ARGS = {
    'input_data': (
        torch.randn(2, 4, 6),
        torch.randint(0, 10, (2, 1)).long(),
    )
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 3, dtype family float, shape (2, 5, 6) — shape (2, 5, 6) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Scheduler Ran
ran on a synthesized input · constructed an object; no result was computed
revsic/torch-diffusion-wavegan/dwg/scheduler.py
code served (permissive licence) · get_code("1c7cc424d9517d3f")
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 an instance of its own class, Scheduler: the run constructed the object and did not call it, so no result was computed. 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
import numpy as np

torch.manual_seed(0)

ARGS = {
    "steps": 10,
    "channels": 4,
    "logit_min": -10.0,
    "logit_max": 10.0,
}
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 Scheduler — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Scheduler 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
TransposedLN Ran
ran on a synthesized input · constructed an object; no result was computed
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("a3b902a89af68b29")
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 an instance of its own class, TransposedLN: the run constructed the object and did not call it, so no result was computed. 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)

class TransposedLN(nn.Module):
    def __init__(self, d):
        super().__init__()
        self.m = nn.Parameter(torch.zeros(1))
        self.s = nn.Parameter(torch.ones(1))

    def forward(self, x):
        s, m = torch.std_mean(x, dim=-2, unbiased=False, keepdim=True)
        y = self.s / s * (x - m + self.m)
        return y

# Create a small instance of TransposedLN with dimension parameter
d = 4
model = TransposedLN(d)

ARGS = {
    "d": d
}
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 TransposedLN — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a TransposedLN 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
calc_diffusion_step_embedding Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
philsyn/diffwave-vocoder/WaveNet.py
code served (permissive licence) · get_code("e56bda7f4e21073b")
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; 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 numpy as np

torch.manual_seed(0)

# Create diffusion_steps as a torch.long tensor with shape (batchsize, 1)
# Using batch size of 2
diffusion_steps = torch.tensor([[1], [5]], dtype=torch.long)

# diffusion_step_embed_dim_in must be an even integer
# Using 4 as a small even dimension
diffusion_step_embed_dim_in = 4

ARGS = {
    "diffusion_steps": diffusion_steps,
    "diffusion_step_embed_dim_in": diffusion_step_embed_dim_in,
}
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 — no output shape is stated: observe_signature.py could not observe it (call:AssertionError)
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
calc_diffusion_step_embedding Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
rf5/diffwave-unconditional/WaveNet.py
code served (permissive licence) · get_code("340fd762c50dbcd3")
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) 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 diffusion_steps tensor with shape (batchsize, 1)
# Using batch size of 2
diffusion_steps = torch.tensor([[5], [10]], dtype=torch.long)

# diffusion_step_embed_dim_in must be even (asserted in function)
diffusion_step_embed_dim_in = 4

ARGS = {
    "diffusion_steps": diffusion_steps,
    "diffusion_step_embed_dim_in": diffusion_step_embed_dim_in,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
dplr Ran
ran; our contract draft was wrong, not the code
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("45091dfbbeb536a4")
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 math
import numpy as np
from einops import rearrange, repeat
import opt_einsum as oe

torch.manual_seed(0)

ARGS = {
    "scaling": "hippo",
    "N": 4,
    "rank": 1,
    "H": 1,
    "dtype": torch.float,
    "real_scale": 1.0,
    "imag_scale": 1.0,
    "random_real": False,
    "random_imag": False,
    "normalize": False,
    "diagonal": True,
    "random_B": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, numpy, opt_einsum, 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
nplr Ran
ran; our contract draft was wrong, not the code
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("e16d05511b7f3b9c")
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
import opt_einsum as oe

torch.manual_seed(0)

ARGS = {
    "measure": "legs",
    "N": 4,
    "rank": 1,
    "dtype": torch.float,
    "diagonalize_precision": True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, opt_einsum, 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
power Ran
ran; our contract draft was wrong, not the code
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("de72ca35554966ec")
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
from einops import rearrange

torch.manual_seed(0)

# Set up parameters for the power function
L = 4  # Power exponent
N = 3  # Matrix dimension
batch_size = 2

# Create a small matrix A with shape (batch_size, N, N)
A = torch.randn(batch_size, N, N)

# Create a vector v with shape (batch_size, N, L)
# L is the sequence length dimension
v = torch.randn(batch_size, N, L)

ARGS = {
    'L': L,
    'A': A,
    'v': v
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, 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
rank_correction Ran
ran, honoured the contract we drafted
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("f442a4d34c7cc88c")
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 (1, 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)

ARGS = {
    'measure': 'legs',
    'N': 4,
    'rank': 1,
    'dtype': torch.float
}
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 (1, 4) — shape (1, 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
transition Ran
ran; our contract draft was wrong, not the code
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("3752f5434bff09e4")
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 numpy as np

# The function takes:
# - measure: str (one of 'legt', 'legs', 'legsd', 'fourier_diag', 'foud', 'fourier', 'fout')
# - N: int (dimension parameter)
# - **measure_args: additional keyword arguments (currently unused in the function body)

# Test with a simple valid case
ARGS = {
    'measure': 'legt',
    'N': 4,
}
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 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
DiffWave Not yet run
unverified (harvested, no recorded run)
revsic/jax-variational-diffwave/vlbdiffwave/diffwave.py
code served (permissive licence) · get_code("c31b8a40c950b42f")
How it ran

no invocation recorded.

DiffWaveBlock Not yet run
unverified (harvested, no recorded run)
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("b7a3bf2ec8acf1a9")
How it ran

no invocation recorded.

OptimModule Not yet run
unverified (harvested, no recorded run)
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("dbc3dbbcec5cea25")
How it ran

no invocation recorded.

S4 Not yet run
unverified (harvested, no recorded run)
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("728c49623235e291")
How it ran

no invocation recorded.

SSKernel Not yet run
unverified (harvested, no recorded run)
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("973900927c12a940")
How it ran

no invocation recorded.

SSKernelDiag Not yet run
unverified (harvested, no recorded run)
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("4ad6f2c5ec72a024")
How it ran

no invocation recorded.

SSKernelNPLR Not yet run
unverified (harvested, no recorded run)
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("2437fe83d525eec9")
How it ran

no invocation recorded.

WaveNet Not yet run
unverified (harvested, no recorded run)
revsic/tf-diffwave/model/wavenet.py
code served (permissive licence) · get_code("2ff412c7141276c0")
How it ran

no invocation recorded.

WaveNetBlock Not yet run
unverified (harvested, no recorded run)
revsic/jax-variational-diffwave/vlbdiffwave/diffwave.py
code served (permissive licence) · get_code("51373087a8e970b5")
How it ran

no invocation recorded.

WaveNet_Speech_Commands Not yet run
unverified (harvested, no recorded run)
philsyn/diffwave-unconditional/WaveNet.py
code served (permissive licence) · get_code("dceb874ae6c13ea1")
How it ran

no invocation recorded.

WaveNet_vocoder Not yet run
unverified (harvested, no recorded run)
philsyn/diffwave-vocoder/WaveNet.py
code served (permissive licence) · get_code("200138f29cbeb7d1")
How it ran

no invocation recorded.

combination Not yet run
unverified (harvested, no recorded run)
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("0853943b994b009b")
How it ran

no invocation recorded.

ssm Not yet run
unverified (harvested, no recorded run)
albertfgu/diffwave-sashimi/models/sashimi.py
code served (permissive licence) · get_code("19f354cdfb484c16")
How it ran

no invocation recorded.

Also present in other papers' code

Of 33 samples lifted from this paper's linked repositories, 33 carry a function identity and 0 do not. Of the 32 distinct functions with an identity, 11 are also present in other papers whose code we hold (7 with at least one paper dated earlier, 0 the same month, 7 later) and 21 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
LinearActivation
brick of this paper's assembly: helper
8 other papers
1 dated earlier, 7 later
1312.5602 (2013-12)
Playing Atari with Deep Reinforcement Learning
2111.00396 (2021-11)
Efficiently Modeling Long Sequences with Structured State Spaces
2211.11176 (2022-11)
Modeling Multivariate Biosignals With Graph Neural Networks and Structured State Space Models
and 5 more not named here
power
brick of this paper's assembly: helper
4 other papers
1 dated earlier, 3 later
1312.5602 (2013-12)
Playing Atari with Deep Reinforcement Learning
2111.00396 (2021-11)
Efficiently Modeling Long Sequences with Structured State Spaces
2204.01692 (2022-04)
Long Movie Clip Classification with State-Space Video Models
and 1 more not named here
calc_diffusion_step_embedding
brick of this paper's assembly: helper
3 other papers
3 later · 1 only as an assembly brick
2203.13508 (2022-03)
BDDM: Bilateral Denoising Diffusion Models for Fast and High-Quality Speech Synthesis
2204.09934 (2022-04)
FastDiff: A Fast Conditional Diffusion Model for High-Quality Speech Synthesis
2306.03258 (2023-06)
LipVoicer: Generating Speech from Silent Videos Guided by Lip Reading
combination
brick of this paper's assembly: helper
3 other papers
1 dated earlier, 2 later
1312.5602 (2013-12)
Playing Atari with Deep Reinforcement Learning
2212.14052 (2022-12)
Hungry Hungry Hippos: Towards Language Modeling with State Space Models
2411.04165 (2024-11)
Bio-xLSTM: Generative modeling, representation and in-context learning of biological and chemical sequences
DiffWave
brick of this paper's assembly: entry
1 other paper
1 later
2107.00630 (2021-07)
Variational Diffusion Models
OptimModule
brick of this paper's assembly: helper
1 other paper
1 dated earlier
1312.5602 (2013-12)
Playing Atari with Deep Reinforcement Learning
WaveNetBlock
brick of this paper's assembly: model
1 other paper
1 later
2107.00630 (2021-07)
Variational Diffusion Models
calc_diffusion_step_embedding
brick of this paper's assembly: helper
1 other paper
1 later
2202.09729 (2022-02)
It's Raw! Audio Generation with State-Space Models
dplr
brick of this paper's assembly: helper
1 other paper
1 dated earlier
1312.5602 (2013-12)
Playing Atari with Deep Reinforcement Learning
nplr
brick of this paper's assembly: helper
1 other paper
1 dated earlier
1312.5602 (2013-12)
Playing Atari with Deep Reinforcement Learning
rank_correction
brick of this paper's assembly: helper
1 other paper
1 dated earlier
1312.5602 (2013-12)
Playing Atari with Deep Reinforcement Learning

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

In this work, we propose DiffWave, a versatile diffusion probabilistic model for conditional and unconditional waveform generation. The model is non-autoregressive, and converts the white noise signal into structured waveform through a Markov chain with a constant number of steps at synthesis. It is efficiently trained by optimizing a variant of variational bound on the data likelihood. DiffWave produces high-fidelity audio in different waveform generation tasks, including neural vocoding conditioned on mel spectrogram, class-conditional generation, and unconditional generation. We demonstrate that DiffWave matches a strong WaveNet vocoder in terms of speech quality (MOS: 4.44 versus 4.43), while synthesizing orders of magnitude faster. In particular, it significantly outperforms autoregressive and GAN-based waveform models in the challenging unconditional generation task in terms of audio quality and sample diversity from various automatic and human evaluations. 1 * Contributed to the work during an internship at Baidu Research, USA.

For agents

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

get_harvested_code_for_paper("2009.09761")
get_code_for_paper("2009.09761")
have("2009.09761")

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

curl https://syntology.ai/api/ran/2009.09761.json

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

[![Syntology run record](https://syntology.ai/api/ran/2009.09761.svg)](https://syntology.ai/paper/2009.09761)

Connect an agent — have() is free.