facebookresearch--audiocraft
Alexandre Défossez (autodoc)
501 行
30 KiB
HTML
501 行
30 KiB
HTML
<!doctype html>
|
||
<html lang="en">
|
||
<head>
|
||
<meta charset="utf-8">
|
||
<meta name="viewport" content="width=device-width, initial-scale=1, minimum-scale=1">
|
||
<meta name="generator" content="pdoc3 0.11.5">
|
||
<title>audiocraft.metrics.kld API documentation</title>
|
||
<meta name="description" content="">
|
||
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/10up-sanitize.css/13.0.0/sanitize.min.css" integrity="sha512-y1dtMcuvtTMJc1yPgEqF0ZjQbhnc/bFhyvIyVNb9Zk5mIGtqVaAB1Ttl28su8AvFMOY0EwRbAe+HCLqj6W7/KA==" crossorigin>
|
||
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/10up-sanitize.css/13.0.0/typography.min.css" integrity="sha512-Y1DYSb995BAfxobCkKepB1BqJJTPrOp3zPL74AWFugHHmmdcvO+C48WLrUOlhGMc0QG7AE3f7gmvvcrmX2fDoA==" crossorigin>
|
||
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.9.0/styles/default.min.css" crossorigin>
|
||
<style>:root{--highlight-color:#fe9}.flex{display:flex !important}body{line-height:1.5em}#content{padding:20px}#sidebar{padding:1.5em;overflow:hidden}#sidebar > *:last-child{margin-bottom:2cm}.http-server-breadcrumbs{font-size:130%;margin:0 0 15px 0}#footer{font-size:.75em;padding:5px 30px;border-top:1px solid #ddd;text-align:right}#footer p{margin:0 0 0 1em;display:inline-block}#footer p:last-child{margin-right:30px}h1,h2,h3,h4,h5{font-weight:300}h1{font-size:2.5em;line-height:1.1em}h2{font-size:1.75em;margin:2em 0 .50em 0}h3{font-size:1.4em;margin:1.6em 0 .7em 0}h4{margin:0;font-size:105%}h1:target,h2:target,h3:target,h4:target,h5:target,h6:target{background:var(--highlight-color);padding:.2em 0}a{color:#058;text-decoration:none;transition:color .2s ease-in-out}a:visited{color:#503}a:hover{color:#b62}.title code{font-weight:bold}h2[id^="header-"]{margin-top:2em}.ident{color:#900;font-weight:bold}pre code{font-size:.8em;line-height:1.4em;padding:1em;display:block}code{background:#f3f3f3;font-family:"DejaVu Sans Mono",monospace;padding:1px 4px;overflow-wrap:break-word}h1 code{background:transparent}pre{border-top:1px solid #ccc;border-bottom:1px solid #ccc;margin:1em 0}#http-server-module-list{display:flex;flex-flow:column}#http-server-module-list div{display:flex}#http-server-module-list dt{min-width:10%}#http-server-module-list p{margin-top:0}.toc ul,#index{list-style-type:none;margin:0;padding:0}#index code{background:transparent}#index h3{border-bottom:1px solid #ddd}#index ul{padding:0}#index h4{margin-top:.6em;font-weight:bold}@media (min-width:200ex){#index .two-column{column-count:2}}@media (min-width:300ex){#index .two-column{column-count:3}}dl{margin-bottom:2em}dl dl:last-child{margin-bottom:4em}dd{margin:0 0 1em 3em}#header-classes + dl > dd{margin-bottom:3em}dd dd{margin-left:2em}dd p{margin:10px 0}.name{background:#eee;font-size:.85em;padding:5px 10px;display:inline-block;min-width:40%}.name:hover{background:#e0e0e0}dt:target .name{background:var(--highlight-color)}.name > span:first-child{white-space:nowrap}.name.class > span:nth-child(2){margin-left:.4em}.inherited{color:#999;border-left:5px solid #eee;padding-left:1em}.inheritance em{font-style:normal;font-weight:bold}.desc h2{font-weight:400;font-size:1.25em}.desc h3{font-size:1em}.desc dt code{background:inherit}.source > summary,.git-link-div{color:#666;text-align:right;font-weight:400;font-size:.8em;text-transform:uppercase}.source summary > *{white-space:nowrap;cursor:pointer}.git-link{color:inherit;margin-left:1em}.source pre{max-height:500px;overflow:auto;margin:0}.source pre code{font-size:12px;overflow:visible;min-width:max-content}.hlist{list-style:none}.hlist li{display:inline}.hlist li:after{content:',\2002'}.hlist li:last-child:after{content:none}.hlist .hlist{display:inline;padding-left:1em}img{max-width:100%}td{padding:0 .5em}.admonition{padding:.1em 1em;margin:1em 0}.admonition-title{font-weight:bold}.admonition.note,.admonition.info,.admonition.important{background:#aef}.admonition.todo,.admonition.versionadded,.admonition.tip,.admonition.hint{background:#dfd}.admonition.warning,.admonition.versionchanged,.admonition.deprecated{background:#fd4}.admonition.error,.admonition.danger,.admonition.caution{background:lightpink}</style>
|
||
<style media="screen and (min-width: 700px)">@media screen and (min-width:700px){#sidebar{width:30%;height:100vh;overflow:auto;position:sticky;top:0}#content{width:70%;max-width:100ch;padding:3em 4em;border-left:1px solid #ddd}pre code{font-size:1em}.name{font-size:1em}main{display:flex;flex-direction:row-reverse;justify-content:flex-end}.toc ul ul,#index ul ul{padding-left:1em}.toc > ul > li{margin-top:.5em}}</style>
|
||
<style media="print">@media print{#sidebar h1{page-break-before:always}.source{display:none}}@media print{*{background:transparent !important;color:#000 !important;box-shadow:none !important;text-shadow:none !important}a[href]:after{content:" (" attr(href) ")";font-size:90%}a[href][title]:after{content:none}abbr[title]:after{content:" (" attr(title) ")"}.ir a:after,a[href^="javascript:"]:after,a[href^="#"]:after{content:""}pre,blockquote{border:1px solid #999;page-break-inside:avoid}thead{display:table-header-group}tr,img{page-break-inside:avoid}img{max-width:100% !important}@page{margin:0.5cm}p,h2,h3{orphans:3;widows:3}h1,h2,h3,h4,h5,h6{page-break-after:avoid}}</style>
|
||
<script defer src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.9.0/highlight.min.js" integrity="sha512-D9gUyxqja7hBtkWpPWGt9wfbfaMGVt9gnyCvYa+jojwwPHLCzUm5i8rpk7vD7wNee9bA35eYIjobYPaQuKS1MQ==" crossorigin></script>
|
||
<script>window.addEventListener('DOMContentLoaded', () => {
|
||
hljs.configure({languages: ['bash', 'css', 'diff', 'graphql', 'ini', 'javascript', 'json', 'plaintext', 'python', 'python-repl', 'rust', 'shell', 'sql', 'typescript', 'xml', 'yaml']});
|
||
hljs.highlightAll();
|
||
/* Collapse source docstrings */
|
||
setTimeout(() => {
|
||
[...document.querySelectorAll('.hljs.language-python > .hljs-string')]
|
||
.filter(el => el.innerHTML.length > 200 && ['"""', "'''"].includes(el.innerHTML.substring(0, 3)))
|
||
.forEach(el => {
|
||
let d = document.createElement('details');
|
||
d.classList.add('hljs-string');
|
||
d.innerHTML = '<summary>"""</summary>' + el.innerHTML.substring(3);
|
||
el.replaceWith(d);
|
||
});
|
||
}, 100);
|
||
})</script>
|
||
</head>
|
||
<body>
|
||
<main>
|
||
<article id="content">
|
||
<header>
|
||
<h1 class="title">Module <code>audiocraft.metrics.kld</code></h1>
|
||
</header>
|
||
<section id="section-intro">
|
||
</section>
|
||
<section>
|
||
</section>
|
||
<section>
|
||
</section>
|
||
<section>
|
||
<h2 class="section-title" id="header-functions">Functions</h2>
|
||
<dl>
|
||
<dt id="audiocraft.metrics.kld.kl_divergence"><code class="name flex">
|
||
<span>def <span class="ident">kl_divergence</span></span>(<span>pred_probs: torch.Tensor, target_probs: torch.Tensor, epsilon: float = 1e-06) ‑> torch.Tensor</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def kl_divergence(pred_probs: torch.Tensor, target_probs: torch.Tensor, epsilon: float = 1e-6) -> torch.Tensor:
|
||
"""Computes the elementwise KL-Divergence loss between probability distributions
|
||
from generated samples and target samples.
|
||
|
||
Args:
|
||
pred_probs (torch.Tensor): Probabilities for each label obtained
|
||
from a classifier on generated audio. Expected shape is [B, num_classes].
|
||
target_probs (torch.Tensor): Probabilities for each label obtained
|
||
from a classifier on target audio. Expected shape is [B, num_classes].
|
||
epsilon (float): Epsilon value.
|
||
Returns:
|
||
kld (torch.Tensor): KLD loss between each generated sample and target pair.
|
||
"""
|
||
kl_div = torch.nn.functional.kl_div((pred_probs + epsilon).log(), target_probs, reduction="none")
|
||
return kl_div.sum(-1)</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Computes the elementwise KL-Divergence loss between probability distributions
|
||
from generated samples and target samples.</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>pred_probs</code></strong> : <code>torch.Tensor</code></dt>
|
||
<dd>Probabilities for each label obtained
|
||
from a classifier on generated audio. Expected shape is [B, num_classes].</dd>
|
||
<dt><strong><code>target_probs</code></strong> : <code>torch.Tensor</code></dt>
|
||
<dd>Probabilities for each label obtained
|
||
from a classifier on target audio. Expected shape is [B, num_classes].</dd>
|
||
<dt><strong><code>epsilon</code></strong> : <code>float</code></dt>
|
||
<dd>Epsilon value.</dd>
|
||
</dl>
|
||
<h2 id="returns">Returns</h2>
|
||
<p>kld (torch.Tensor): KLD loss between each generated sample and target pair.</p></div>
|
||
</dd>
|
||
</dl>
|
||
</section>
|
||
<section>
|
||
<h2 class="section-title" id="header-classes">Classes</h2>
|
||
<dl>
|
||
<dt id="audiocraft.metrics.kld.KLDivergenceMetric"><code class="flex name class">
|
||
<span>class <span class="ident">KLDivergenceMetric</span></span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">class KLDivergenceMetric(torchmetrics.Metric):
|
||
"""Base implementation for KL Divergence metric.
|
||
|
||
The KL divergence is measured between probability distributions
|
||
of class predictions returned by a pre-trained audio classification model.
|
||
When the KL-divergence is low, the generated audio is expected to
|
||
have similar acoustic characteristics as the reference audio,
|
||
according to the classifier.
|
||
"""
|
||
def __init__(self):
|
||
super().__init__()
|
||
self.add_state("kld_pq_sum", default=torch.tensor(0.), dist_reduce_fx="sum")
|
||
self.add_state("kld_qp_sum", default=torch.tensor(0.), dist_reduce_fx="sum")
|
||
self.add_state("kld_all_sum", default=torch.tensor(0.), dist_reduce_fx="sum")
|
||
self.add_state("weight", default=torch.tensor(0), dist_reduce_fx="sum")
|
||
|
||
def _get_label_distribution(self, x: torch.Tensor, sizes: torch.Tensor,
|
||
sample_rates: torch.Tensor) -> tp.Optional[torch.Tensor]:
|
||
"""Get model output given provided input tensor.
|
||
|
||
Args:
|
||
x (torch.Tensor): Input audio tensor of shape [B, C, T].
|
||
sizes (torch.Tensor): Actual audio sample length, of shape [B].
|
||
sample_rates (torch.Tensor): Actual audio sample rate, of shape [B].
|
||
Returns:
|
||
probs (torch.Tensor): Probabilities over labels, of shape [B, num_classes].
|
||
"""
|
||
raise NotImplementedError("implement method to extract label distributions from the model.")
|
||
|
||
def update(self, preds: torch.Tensor, targets: torch.Tensor,
|
||
sizes: torch.Tensor, sample_rates: torch.Tensor) -> None:
|
||
"""Calculates running KL-Divergence loss between batches of audio
|
||
preds (generated) and target (ground-truth)
|
||
Args:
|
||
preds (torch.Tensor): Audio samples to evaluate, of shape [B, C, T].
|
||
targets (torch.Tensor): Target samples to compare against, of shape [B, C, T].
|
||
sizes (torch.Tensor): Actual audio sample length, of shape [B].
|
||
sample_rates (torch.Tensor): Actual audio sample rate, of shape [B].
|
||
"""
|
||
assert preds.shape == targets.shape
|
||
assert preds.size(0) > 0, "Cannot update the loss with empty tensors"
|
||
preds_probs = self._get_label_distribution(preds, sizes, sample_rates)
|
||
targets_probs = self._get_label_distribution(targets, sizes, sample_rates)
|
||
if preds_probs is not None and targets_probs is not None:
|
||
assert preds_probs.shape == targets_probs.shape
|
||
kld_scores = kl_divergence(preds_probs, targets_probs)
|
||
assert not torch.isnan(kld_scores).any(), "kld_scores contains NaN value(s)!"
|
||
self.kld_pq_sum += torch.sum(kld_scores)
|
||
kld_qp_scores = kl_divergence(targets_probs, preds_probs)
|
||
self.kld_qp_sum += torch.sum(kld_qp_scores)
|
||
self.weight += torch.tensor(kld_scores.size(0))
|
||
|
||
def compute(self) -> dict:
|
||
"""Computes KL-Divergence across all evaluated pred/target pairs."""
|
||
weight: float = float(self.weight.item()) # type: ignore
|
||
assert weight > 0, "Unable to compute with total number of comparisons <= 0"
|
||
logger.info(f"Computing KL divergence on a total of {weight} samples")
|
||
kld_pq = self.kld_pq_sum.item() / weight # type: ignore
|
||
kld_qp = self.kld_qp_sum.item() / weight # type: ignore
|
||
kld_both = kld_pq + kld_qp
|
||
return {'kld': kld_pq, 'kld_pq': kld_pq, 'kld_qp': kld_qp, 'kld_both': kld_both}</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Base implementation for KL Divergence metric.</p>
|
||
<p>The KL divergence is measured between probability distributions
|
||
of class predictions returned by a pre-trained audio classification model.
|
||
When the KL-divergence is low, the generated audio is expected to
|
||
have similar acoustic characteristics as the reference audio,
|
||
according to the classifier.</p>
|
||
<p>Initializes internal Module state, shared by both nn.Module and ScriptModule.</p></div>
|
||
<h3>Ancestors</h3>
|
||
<ul class="hlist">
|
||
<li>torchmetrics.metric.Metric</li>
|
||
<li>torch.nn.modules.module.Module</li>
|
||
<li>abc.ABC</li>
|
||
</ul>
|
||
<h3>Subclasses</h3>
|
||
<ul class="hlist">
|
||
<li><a title="audiocraft.metrics.kld.PasstKLDivergenceMetric" href="#audiocraft.metrics.kld.PasstKLDivergenceMetric">PasstKLDivergenceMetric</a></li>
|
||
</ul>
|
||
<h3>Class variables</h3>
|
||
<dl>
|
||
<dt id="audiocraft.metrics.kld.KLDivergenceMetric.full_state_update"><code class="name">var <span class="ident">full_state_update</span> : bool | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.KLDivergenceMetric.higher_is_better"><code class="name">var <span class="ident">higher_is_better</span> : bool | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.KLDivergenceMetric.is_differentiable"><code class="name">var <span class="ident">is_differentiable</span> : bool | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.KLDivergenceMetric.plot_legend_name"><code class="name">var <span class="ident">plot_legend_name</span> : str | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.KLDivergenceMetric.plot_lower_bound"><code class="name">var <span class="ident">plot_lower_bound</span> : float | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.KLDivergenceMetric.plot_upper_bound"><code class="name">var <span class="ident">plot_upper_bound</span> : float | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
</dl>
|
||
<h3>Methods</h3>
|
||
<dl>
|
||
<dt id="audiocraft.metrics.kld.KLDivergenceMetric.compute"><code class="name flex">
|
||
<span>def <span class="ident">compute</span></span>(<span>self) ‑> dict</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def compute(self) -> dict:
|
||
"""Computes KL-Divergence across all evaluated pred/target pairs."""
|
||
weight: float = float(self.weight.item()) # type: ignore
|
||
assert weight > 0, "Unable to compute with total number of comparisons <= 0"
|
||
logger.info(f"Computing KL divergence on a total of {weight} samples")
|
||
kld_pq = self.kld_pq_sum.item() / weight # type: ignore
|
||
kld_qp = self.kld_qp_sum.item() / weight # type: ignore
|
||
kld_both = kld_pq + kld_qp
|
||
return {'kld': kld_pq, 'kld_pq': kld_pq, 'kld_qp': kld_qp, 'kld_both': kld_both}</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Computes KL-Divergence across all evaluated pred/target pairs.</p></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.KLDivergenceMetric.update"><code class="name flex">
|
||
<span>def <span class="ident">update</span></span>(<span>self,<br>preds: torch.Tensor,<br>targets: torch.Tensor,<br>sizes: torch.Tensor,<br>sample_rates: torch.Tensor) ‑> None</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">def update(self, preds: torch.Tensor, targets: torch.Tensor,
|
||
sizes: torch.Tensor, sample_rates: torch.Tensor) -> None:
|
||
"""Calculates running KL-Divergence loss between batches of audio
|
||
preds (generated) and target (ground-truth)
|
||
Args:
|
||
preds (torch.Tensor): Audio samples to evaluate, of shape [B, C, T].
|
||
targets (torch.Tensor): Target samples to compare against, of shape [B, C, T].
|
||
sizes (torch.Tensor): Actual audio sample length, of shape [B].
|
||
sample_rates (torch.Tensor): Actual audio sample rate, of shape [B].
|
||
"""
|
||
assert preds.shape == targets.shape
|
||
assert preds.size(0) > 0, "Cannot update the loss with empty tensors"
|
||
preds_probs = self._get_label_distribution(preds, sizes, sample_rates)
|
||
targets_probs = self._get_label_distribution(targets, sizes, sample_rates)
|
||
if preds_probs is not None and targets_probs is not None:
|
||
assert preds_probs.shape == targets_probs.shape
|
||
kld_scores = kl_divergence(preds_probs, targets_probs)
|
||
assert not torch.isnan(kld_scores).any(), "kld_scores contains NaN value(s)!"
|
||
self.kld_pq_sum += torch.sum(kld_scores)
|
||
kld_qp_scores = kl_divergence(targets_probs, preds_probs)
|
||
self.kld_qp_sum += torch.sum(kld_qp_scores)
|
||
self.weight += torch.tensor(kld_scores.size(0))</code></pre>
|
||
</details>
|
||
<div class="desc"><p>Calculates running KL-Divergence loss between batches of audio
|
||
preds (generated) and target (ground-truth)</p>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>preds</code></strong> : <code>torch.Tensor</code></dt>
|
||
<dd>Audio samples to evaluate, of shape [B, C, T].</dd>
|
||
<dt><strong><code>targets</code></strong> : <code>torch.Tensor</code></dt>
|
||
<dd>Target samples to compare against, of shape [B, C, T].</dd>
|
||
<dt><strong><code>sizes</code></strong> : <code>torch.Tensor</code></dt>
|
||
<dd>Actual audio sample length, of shape [B].</dd>
|
||
<dt><strong><code>sample_rates</code></strong> : <code>torch.Tensor</code></dt>
|
||
<dd>Actual audio sample rate, of shape [B].</dd>
|
||
</dl></div>
|
||
</dd>
|
||
</dl>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.PasstKLDivergenceMetric"><code class="flex name class">
|
||
<span>class <span class="ident">PasstKLDivergenceMetric</span></span>
|
||
<span>(</span><span>pretrained_length: float | None = None)</span>
|
||
</code></dt>
|
||
<dd>
|
||
<details class="source">
|
||
<summary>
|
||
<span>Expand source code</span>
|
||
</summary>
|
||
<pre><code class="python">class PasstKLDivergenceMetric(KLDivergenceMetric):
|
||
"""KL-Divergence metric based on pre-trained PASST classifier on AudioSet.
|
||
|
||
From: PaSST: Efficient Training of Audio Transformers with Patchout
|
||
Paper: https://arxiv.org/abs/2110.05069
|
||
Implementation: https://github.com/kkoutini/PaSST
|
||
|
||
Follow instructions from the github repo:
|
||
```
|
||
pip install 'git+https://github.com/kkoutini/passt_hear21@0.0.19#egg=hear21passt'
|
||
```
|
||
|
||
Args:
|
||
pretrained_length (float, optional): Audio duration used for the pretrained model.
|
||
"""
|
||
def __init__(self, pretrained_length: tp.Optional[float] = None):
|
||
super().__init__()
|
||
self._initialize_model(pretrained_length)
|
||
|
||
def _initialize_model(self, pretrained_length: tp.Optional[float] = None):
|
||
"""Initialize underlying PaSST audio classifier."""
|
||
model, sr, max_frames, min_frames = self._load_base_model(pretrained_length)
|
||
self.min_input_frames = min_frames
|
||
self.max_input_frames = max_frames
|
||
self.model_sample_rate = sr
|
||
self.model = model
|
||
self.model.eval()
|
||
self.model.to(self.device)
|
||
|
||
def _load_base_model(self, pretrained_length: tp.Optional[float]):
|
||
"""Load pretrained model from PaSST."""
|
||
try:
|
||
if pretrained_length == 30:
|
||
from hear21passt.base30sec import get_basic_model # type: ignore
|
||
max_duration = 30
|
||
elif pretrained_length == 20:
|
||
from hear21passt.base20sec import get_basic_model # type: ignore
|
||
max_duration = 20
|
||
else:
|
||
from hear21passt.base import get_basic_model # type: ignore
|
||
# Original PASST was trained on AudioSet with 10s-long audio samples
|
||
max_duration = 10
|
||
min_duration = 0.15
|
||
min_duration = 0.15
|
||
except ModuleNotFoundError:
|
||
raise ModuleNotFoundError(
|
||
"Please install hear21passt to compute KL divergence: ",
|
||
"pip install 'git+https://github.com/kkoutini/passt_hear21@0.0.19#egg=hear21passt'"
|
||
)
|
||
model_sample_rate = 32_000
|
||
max_input_frames = int(max_duration * model_sample_rate)
|
||
min_input_frames = int(min_duration * model_sample_rate)
|
||
with open(os.devnull, 'w') as f, contextlib.redirect_stdout(f):
|
||
model = get_basic_model(mode='logits')
|
||
return model, model_sample_rate, max_input_frames, min_input_frames
|
||
|
||
def _process_audio(self, wav: torch.Tensor, sample_rate: int, wav_len: int) -> tp.List[torch.Tensor]:
|
||
"""Process audio to feed to the pretrained model."""
|
||
wav = wav.unsqueeze(0)
|
||
wav = wav[..., :wav_len]
|
||
wav = convert_audio(wav, from_rate=sample_rate, to_rate=self.model_sample_rate, to_channels=1)
|
||
wav = wav.squeeze(0)
|
||
# we don't pad but return a list of audio segments as this otherwise affects the KLD computation
|
||
segments = torch.split(wav, self.max_input_frames, dim=-1)
|
||
valid_segments = []
|
||
for s in segments:
|
||
# ignoring too small segments that are breaking the model inference
|
||
if s.size(-1) > self.min_input_frames:
|
||
valid_segments.append(s)
|
||
return [s[None] for s in valid_segments]
|
||
|
||
def _get_model_preds(self, wav: torch.Tensor) -> torch.Tensor:
|
||
"""Run the pretrained model and get the predictions."""
|
||
assert wav.dim() == 3, f"Unexpected number of dims for preprocessed wav: {wav.shape}"
|
||
wav = wav.mean(dim=1)
|
||
# PaSST is printing a lot of garbage that we are not interested in
|
||
with open(os.devnull, "w") as f, contextlib.redirect_stdout(f):
|
||
with torch.no_grad(), _patch_passt_stft():
|
||
logits = self.model(wav.to(self.device))
|
||
probs = torch.softmax(logits, dim=-1)
|
||
return probs
|
||
|
||
def _get_label_distribution(self, x: torch.Tensor, sizes: torch.Tensor,
|
||
sample_rates: torch.Tensor) -> tp.Optional[torch.Tensor]:
|
||
"""Get model output given provided input tensor.
|
||
|
||
Args:
|
||
x (torch.Tensor): Input audio tensor of shape [B, C, T].
|
||
sizes (torch.Tensor): Actual audio sample length, of shape [B].
|
||
sample_rates (torch.Tensor): Actual audio sample rate, of shape [B].
|
||
Returns:
|
||
probs (torch.Tensor, optional): Probabilities over labels, of shape [B, num_classes].
|
||
"""
|
||
all_probs: tp.List[torch.Tensor] = []
|
||
for i, wav in enumerate(x):
|
||
sample_rate = int(sample_rates[i].item())
|
||
wav_len = int(sizes[i].item())
|
||
wav_segments = self._process_audio(wav, sample_rate, wav_len)
|
||
for segment in wav_segments:
|
||
probs = self._get_model_preds(segment).mean(dim=0)
|
||
all_probs.append(probs)
|
||
if len(all_probs) > 0:
|
||
return torch.stack(all_probs, dim=0)
|
||
else:
|
||
return None</code></pre>
|
||
</details>
|
||
<div class="desc"><p>KL-Divergence metric based on pre-trained PASST classifier on AudioSet.</p>
|
||
<p>From: PaSST: Efficient Training of Audio Transformers with Patchout
|
||
Paper: <a href="https://arxiv.org/abs/2110.05069">https://arxiv.org/abs/2110.05069</a>
|
||
Implementation: <a href="https://github.com/kkoutini/PaSST">https://github.com/kkoutini/PaSST</a></p>
|
||
<p>Follow instructions from the github repo:</p>
|
||
<pre><code>pip install 'git+https://github.com/kkoutini/passt_hear21@0.0.19#egg=hear21passt'
|
||
</code></pre>
|
||
<h2 id="args">Args</h2>
|
||
<dl>
|
||
<dt><strong><code>pretrained_length</code></strong> : <code>float</code>, optional</dt>
|
||
<dd>Audio duration used for the pretrained model.</dd>
|
||
</dl>
|
||
<p>Initializes internal Module state, shared by both nn.Module and ScriptModule.</p></div>
|
||
<h3>Ancestors</h3>
|
||
<ul class="hlist">
|
||
<li><a title="audiocraft.metrics.kld.KLDivergenceMetric" href="#audiocraft.metrics.kld.KLDivergenceMetric">KLDivergenceMetric</a></li>
|
||
<li>torchmetrics.metric.Metric</li>
|
||
<li>torch.nn.modules.module.Module</li>
|
||
<li>abc.ABC</li>
|
||
</ul>
|
||
<h3>Class variables</h3>
|
||
<dl>
|
||
<dt id="audiocraft.metrics.kld.PasstKLDivergenceMetric.full_state_update"><code class="name">var <span class="ident">full_state_update</span> : bool | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.PasstKLDivergenceMetric.higher_is_better"><code class="name">var <span class="ident">higher_is_better</span> : bool | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.PasstKLDivergenceMetric.is_differentiable"><code class="name">var <span class="ident">is_differentiable</span> : bool | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.PasstKLDivergenceMetric.plot_legend_name"><code class="name">var <span class="ident">plot_legend_name</span> : str | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.PasstKLDivergenceMetric.plot_lower_bound"><code class="name">var <span class="ident">plot_lower_bound</span> : float | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
<dt id="audiocraft.metrics.kld.PasstKLDivergenceMetric.plot_upper_bound"><code class="name">var <span class="ident">plot_upper_bound</span> : float | None</code></dt>
|
||
<dd>
|
||
<div class="desc"></div>
|
||
</dd>
|
||
</dl>
|
||
<h3>Inherited members</h3>
|
||
<ul class="hlist">
|
||
<li><code><b><a title="audiocraft.metrics.kld.KLDivergenceMetric" href="#audiocraft.metrics.kld.KLDivergenceMetric">KLDivergenceMetric</a></b></code>:
|
||
<ul class="hlist">
|
||
<li><code><a title="audiocraft.metrics.kld.KLDivergenceMetric.compute" href="#audiocraft.metrics.kld.KLDivergenceMetric.compute">compute</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.KLDivergenceMetric.update" href="#audiocraft.metrics.kld.KLDivergenceMetric.update">update</a></code></li>
|
||
</ul>
|
||
</li>
|
||
</ul>
|
||
</dd>
|
||
</dl>
|
||
</section>
|
||
</article>
|
||
<nav id="sidebar">
|
||
<div class="toc">
|
||
<ul></ul>
|
||
</div>
|
||
<ul id="index">
|
||
<li><h3>Super-module</h3>
|
||
<ul>
|
||
<li><code><a title="audiocraft.metrics" href="index.html">audiocraft.metrics</a></code></li>
|
||
</ul>
|
||
</li>
|
||
<li><h3><a href="#header-functions">Functions</a></h3>
|
||
<ul class="">
|
||
<li><code><a title="audiocraft.metrics.kld.kl_divergence" href="#audiocraft.metrics.kld.kl_divergence">kl_divergence</a></code></li>
|
||
</ul>
|
||
</li>
|
||
<li><h3><a href="#header-classes">Classes</a></h3>
|
||
<ul>
|
||
<li>
|
||
<h4><code><a title="audiocraft.metrics.kld.KLDivergenceMetric" href="#audiocraft.metrics.kld.KLDivergenceMetric">KLDivergenceMetric</a></code></h4>
|
||
<ul class="two-column">
|
||
<li><code><a title="audiocraft.metrics.kld.KLDivergenceMetric.compute" href="#audiocraft.metrics.kld.KLDivergenceMetric.compute">compute</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.KLDivergenceMetric.full_state_update" href="#audiocraft.metrics.kld.KLDivergenceMetric.full_state_update">full_state_update</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.KLDivergenceMetric.higher_is_better" href="#audiocraft.metrics.kld.KLDivergenceMetric.higher_is_better">higher_is_better</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.KLDivergenceMetric.is_differentiable" href="#audiocraft.metrics.kld.KLDivergenceMetric.is_differentiable">is_differentiable</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.KLDivergenceMetric.plot_legend_name" href="#audiocraft.metrics.kld.KLDivergenceMetric.plot_legend_name">plot_legend_name</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.KLDivergenceMetric.plot_lower_bound" href="#audiocraft.metrics.kld.KLDivergenceMetric.plot_lower_bound">plot_lower_bound</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.KLDivergenceMetric.plot_upper_bound" href="#audiocraft.metrics.kld.KLDivergenceMetric.plot_upper_bound">plot_upper_bound</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.KLDivergenceMetric.update" href="#audiocraft.metrics.kld.KLDivergenceMetric.update">update</a></code></li>
|
||
</ul>
|
||
</li>
|
||
<li>
|
||
<h4><code><a title="audiocraft.metrics.kld.PasstKLDivergenceMetric" href="#audiocraft.metrics.kld.PasstKLDivergenceMetric">PasstKLDivergenceMetric</a></code></h4>
|
||
<ul class="two-column">
|
||
<li><code><a title="audiocraft.metrics.kld.PasstKLDivergenceMetric.full_state_update" href="#audiocraft.metrics.kld.PasstKLDivergenceMetric.full_state_update">full_state_update</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.PasstKLDivergenceMetric.higher_is_better" href="#audiocraft.metrics.kld.PasstKLDivergenceMetric.higher_is_better">higher_is_better</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.PasstKLDivergenceMetric.is_differentiable" href="#audiocraft.metrics.kld.PasstKLDivergenceMetric.is_differentiable">is_differentiable</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.PasstKLDivergenceMetric.plot_legend_name" href="#audiocraft.metrics.kld.PasstKLDivergenceMetric.plot_legend_name">plot_legend_name</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.PasstKLDivergenceMetric.plot_lower_bound" href="#audiocraft.metrics.kld.PasstKLDivergenceMetric.plot_lower_bound">plot_lower_bound</a></code></li>
|
||
<li><code><a title="audiocraft.metrics.kld.PasstKLDivergenceMetric.plot_upper_bound" href="#audiocraft.metrics.kld.PasstKLDivergenceMetric.plot_upper_bound">plot_upper_bound</a></code></li>
|
||
</ul>
|
||
</li>
|
||
</ul>
|
||
</li>
|
||
</ul>
|
||
</nav>
|
||
</main>
|
||
<footer id="footer">
|
||
<p>Generated by <a href="https://pdoc3.github.io/pdoc" title="pdoc: Python API documentation generator"><cite>pdoc</cite> 0.11.5</a>.</p>
|
||
</footer>
|
||
</body>
|
||
</html>
|