Transformers documentation

Audio Spectrogram Transformer

Transformers

You are viewing main version, which requires installation from source. If you'd like regular pip install, checkout the latest stable version (v4.47.1).

Join the Hugging Face community

and get access to the augmented documentation experience

Collaborate on models, datasets and Spaces

Faster examples with accelerated inference

Switch between documentation themes

to get started

Audio Spectrogram Transformer

Overview

The Audio Spectrogram Transformer model was proposed in AST: Audio Spectrogram Transformer by Yuan Gong, Yu-An Chung, James Glass. The Audio Spectrogram Transformer applies a Vision Transformer to audio, by turning audio into an image (spectrogram). The model obtains state-of-the-art results for audio classification.

The abstract from the paper is the following:

In the past decade, convolutional neural networks (CNNs) have been widely adopted as the main building block for end-to-end audio classification models, which aim to learn a direct mapping from audio spectrograms to corresponding labels. To better capture long-range global context, a recent trend is to add a self-attention mechanism on top of the CNN, forming a CNN-attention hybrid model. However, it is unclear whether the reliance on a CNN is necessary, and if neural networks purely based on attention are sufficient to obtain good performance in audio classification. In this paper, we answer the question by introducing the Audio Spectrogram Transformer (AST), the first convolution-free, purely attention-based model for audio classification. We evaluate AST on various audio classification benchmarks, where it achieves new state-of-the-art results of 0.485 mAP on AudioSet, 95.6% accuracy on ESC-50, and 98.1% accuracy on Speech Commands V2.

Audio Spectrogram Transformer architecture. Taken from the original paper.

This model was contributed by nielsr. The original code can be found here.

Usage tips

When fine-tuning the Audio Spectrogram Transformer (AST) on your own dataset, it’s recommended to take care of the input normalization (to make sure the input has mean of 0 and std of 0.5). ASTFeatureExtractor takes care of this. Note that it uses the AudioSet mean and std by default. You can check ast/src/get_norm_stats.py to see how the authors compute the stats for a downstream dataset.
Note that the AST needs a low learning rate (the authors use a 10 times smaller learning rate compared to their CNN model proposed in the PSLA paper) and converges quickly, so please search for a suitable learning rate and learning rate scheduler for your task.

Using Scaled Dot Product Attention (SDPA)

PyTorch includes a native scaled dot-product attention (SDPA) operator as part of torch.nn.functional. This function encompasses several implementations that can be applied depending on the inputs and the hardware in use. See the official documentation or the GPU Inference page for more information.

SDPA is used by default for torch>=2.1.1 when an implementation is available, but you may also set attn_implementation="sdpa" in from_pretrained() to explicitly request SDPA to be used.

from transformers import ASTForAudioClassification
model = ASTForAudioClassification.from_pretrained("MIT/ast-finetuned-audioset-10-10-0.4593", attn_implementation="sdpa", torch_dtype=torch.float16)
...

For the best speedups, we recommend loading the model in half-precision (e.g. torch.float16 or torch.bfloat16).

On a local benchmark (A100-40GB, PyTorch 2.3.0, OS Ubuntu 22.04) with float32 and MIT/ast-finetuned-audioset-10-10-0.4593 model, we saw the following speedups during inference.

Batch size	Average inference time (ms), eager mode	Average inference time (ms), sdpa model	Speed up, Sdpa / Eager (x)
1	27	6	4.5
2	12	6	2
4	21	8	2.62
8	40	14	2.86

Resources

A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with the Audio Spectrogram Transformer.

Audio Classification

A notebook illustrating inference with AST for audio classification can be found here.
ASTForAudioClassification is supported by this example script and notebook.
See also: Audio classification.

If you’re interested in submitting a resource to be included here, please feel free to open a Pull Request and we’ll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.

ASTConfig

class transformers.ASTConfig

< source >

( hidden_size = 768 num_hidden_layers = 12 num_attention_heads = 12 intermediate_size = 3072 hidden_act = 'gelu' hidden_dropout_prob = 0.0 attention_probs_dropout_prob = 0.0 initializer_range = 0.02 layer_norm_eps = 1e-12 patch_size = 16 qkv_bias = True frequency_stride = 10 time_stride = 10 max_length = 1024 num_mel_bins = 128 **kwargs )

Parameters

hidden_size (int, optional, defaults to 768) — Dimensionality of the encoder layers and the pooler layer.
num_hidden_layers (int, optional, defaults to 12) — Number of hidden layers in the Transformer encoder.
num_attention_heads (int, optional, defaults to 12) — Number of attention heads for each attention layer in the Transformer encoder.
intermediate_size (int, optional, defaults to 3072) — Dimensionality of the “intermediate” (i.e., feed-forward) layer in the Transformer encoder.
hidden_act (str or function, optional, defaults to "gelu") — The non-linear activation function (function or string) in the encoder and pooler. If string, "gelu", "relu", "selu" and "gelu_new" are supported.
hidden_dropout_prob (float, optional, defaults to 0.0) — The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
attention_probs_dropout_prob (float, optional, defaults to 0.0) — The dropout ratio for the attention probabilities.
initializer_range (float, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
layer_norm_eps (float, optional, defaults to 1e-12) — The epsilon used by the layer normalization layers.
patch_size (int, optional, defaults to 16) — The size (resolution) of each patch.
qkv_bias (bool, optional, defaults to True) — Whether to add a bias to the queries, keys and values.
frequency_stride (int, optional, defaults to 10) — Frequency stride to use when patchifying the spectrograms.
time_stride (int, optional, defaults to 10) — Temporal stride to use when patchifying the spectrograms.
max_length (int, optional, defaults to 1024) — Temporal dimension of the spectrograms.
num_mel_bins (int, optional, defaults to 128) — Frequency dimension of the spectrograms (number of Mel-frequency bins).

This is the configuration class to store the configuration of a ASTModel. It is used to instantiate an AST model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the AST MIT/ast-finetuned-audioset-10-10-0.4593 architecture.

Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.

Example:

>>> from transformers import ASTConfig, ASTModel

>>> # Initializing a AST MIT/ast-finetuned-audioset-10-10-0.4593 style configuration
>>> configuration = ASTConfig()

>>> # Initializing a model (with random weights) from the MIT/ast-finetuned-audioset-10-10-0.4593 style configuration
>>> model = ASTModel(configuration)

>>> # Accessing the model configuration
>>> configuration = model.config

ASTFeatureExtractor

class transformers.ASTFeatureExtractor

< source >

( feature_size = 1 sampling_rate = 16000 num_mel_bins = 128 max_length = 1024 padding_value = 0.0 do_normalize = True mean = -4.2677393 std = 4.5689974 return_attention_mask = False **kwargs )

Parameters

feature_size (int, optional, defaults to 1) — The feature dimension of the extracted features.
sampling_rate (int, optional, defaults to 16000) — The sampling rate at which the audio files should be digitalized expressed in hertz (Hz).
num_mel_bins (int, optional, defaults to 128) — Number of Mel-frequency bins.
max_length (int, optional, defaults to 1024) — Maximum length to which to pad/truncate the extracted features.
do_normalize (bool, optional, defaults to True) — Whether or not to normalize the log-Mel features using mean and std.
mean (float, optional, defaults to -4.2677393) — The mean value used to normalize the log-Mel features. Uses the AudioSet mean by default.
std (float, optional, defaults to 4.5689974) — The standard deviation value used to normalize the log-Mel features. Uses the AudioSet standard deviation by default.
return_attention_mask (bool, optional, defaults to False) — Whether or not call() should return attention_mask.

Constructs a Audio Spectrogram Transformer (AST) feature extractor.

This feature extractor inherits from SequenceFeatureExtractor which contains most of the main methods. Users should refer to this superclass for more information regarding those methods.

This class extracts mel-filter bank features from raw speech using TorchAudio if installed or using numpy otherwise, pads/truncates them to a fixed length and normalizes them using a mean and standard deviation.

call

< source >

( raw_speech: typing.Union[numpy.ndarray, typing.List[float], typing.List[numpy.ndarray], typing.List[typing.List[float]]] sampling_rate: typing.Optional[int] = None return_tensors: typing.Union[str, transformers.utils.generic.TensorType, NoneType] = None **kwargs )

Parameters

raw_speech (np.ndarray, List[float], List[np.ndarray], List[List[float]]) — The sequence or batch of sequences to be padded. Each sequence can be a numpy array, a list of float values, a list of numpy arrays or a list of list of float values. Must be mono channel audio, not stereo, i.e. single float per timestep.
sampling_rate (int, optional) — The sampling rate at which the raw_speech input was sampled. It is strongly recommended to pass sampling_rate at the forward call to prevent silent errors.
return_tensors (str or TensorType, optional) — If set, will return tensors instead of list of python integers. Acceptable values are:
- 'tf': Return TensorFlow tf.constant objects.
- 'pt': Return PyTorch torch.Tensor objects.
- 'np': Return Numpy np.ndarray objects.

Main method to featurize and prepare for the model one or several sequence(s).

ASTModel

class transformers.ASTModel

< source >

( config: ASTConfig )

Parameters

config (ASTConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

The bare AST Model transformer outputting raw hidden-states without any specific head on top. This model is a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

forward

< source >

( input_values: typing.Optional[torch.Tensor] = None head_mask: typing.Optional[torch.Tensor] = None output_attentions: typing.Optional[bool] = None output_hidden_states: typing.Optional[bool] = None return_dict: typing.Optional[bool] = None ) → transformers.modeling_outputs.BaseModelOutputWithPooling or tuple(torch.FloatTensor)

Parameters

input_values (torch.FloatTensor of shape (batch_size, max_length, num_mel_bins)) — Float values mel features extracted from the raw audio waveform. Raw audio waveform can be obtained by loading a .flac or .wav audio file into an array of type List[float] or a numpy.ndarray, e.g. via the soundfile library (pip install soundfile). To prepare the array into input_features, the AutoFeatureExtractor should be used for extracting the mel features, padding and conversion into a tensor of type torch.FloatTensor. See call()
head_mask (torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) — Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]:
- 1 indicates the head is not masked,
- 0 indicates the head is masked.
output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
return_dict (bool, optional) — Whether or not to return a ModelOutput instead of a plain tuple.

Returns

transformers.modeling_outputs.BaseModelOutputWithPooling or tuple(torch.FloatTensor)

A transformers.modeling_outputs.BaseModelOutputWithPooling or a tuple of torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various elements depending on the configuration (ASTConfig) and inputs.

last_hidden_state (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.
pooler_output (torch.FloatTensor of shape (batch_size, hidden_size)) — Last layer hidden-state of the first token of the sequence (classification token) after further processing through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns the classification token after processing through a linear layer and a tanh activation function. The linear layer weights are trained from the next sentence prediction (classification) objective during pretraining.
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.FloatTensor (one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

The ASTModel forward method, overrides the __call__ special method.

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Example:

>>> from transformers import AutoProcessor, ASTModel
>>> import torch
>>> from datasets import load_dataset

>>> dataset = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation", trust_remote_code=True)
>>> dataset = dataset.sort("id")
>>> sampling_rate = dataset.features["audio"].sampling_rate

>>> processor = AutoProcessor.from_pretrained("MIT/ast-finetuned-audioset-10-10-0.4593")
>>> model = ASTModel.from_pretrained("MIT/ast-finetuned-audioset-10-10-0.4593")

>>> # audio file is decoded on the fly
>>> inputs = processor(dataset[0]["audio"]["array"], sampling_rate=sampling_rate, return_tensors="pt")
>>> with torch.no_grad():
...     outputs = model(**inputs)

>>> last_hidden_states = outputs.last_hidden_state
>>> list(last_hidden_states.shape)
[1, 1214, 768]

ASTForAudioClassification

class transformers.ASTForAudioClassification

< source >

( config: ASTConfig )

Parameters

config (ASTConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

Audio Spectrogram Transformer model with an audio classification head on top (a linear layer on top of the pooled output) e.g. for datasets like AudioSet, Speech Commands v2.

This model is a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

forward

< source >

( input_values: typing.Optional[torch.Tensor] = None head_mask: typing.Optional[torch.Tensor] = None labels: typing.Optional[torch.Tensor] = None output_attentions: typing.Optional[bool] = None output_hidden_states: typing.Optional[bool] = None return_dict: typing.Optional[bool] = None ) → transformers.modeling_outputs.SequenceClassifierOutput or tuple(torch.FloatTensor)

Parameters

input_values (torch.FloatTensor of shape (batch_size, max_length, num_mel_bins)) — Float values mel features extracted from the raw audio waveform. Raw audio waveform can be obtained by loading a .flac or .wav audio file into an array of type List[float] or a numpy.ndarray, e.g. via the soundfile library (pip install soundfile). To prepare the array into input_features, the AutoFeatureExtractor should be used for extracting the mel features, padding and conversion into a tensor of type torch.FloatTensor. See call()
head_mask (torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) — Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]:
- 1 indicates the head is not masked,
- 0 indicates the head is masked.
output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
return_dict (bool, optional) — Whether or not to return a ModelOutput instead of a plain tuple.
labels (torch.LongTensor of shape (batch_size,), optional) — Labels for computing the audio classification/regression loss. Indices should be in [0, ..., config.num_labels - 1]. If config.num_labels == 1 a regression loss is computed (Mean-Square loss), If config.num_labels > 1 a classification loss is computed (Cross-Entropy).

Returns

transformers.modeling_outputs.SequenceClassifierOutput or tuple(torch.FloatTensor)

A transformers.modeling_outputs.SequenceClassifierOutput or a tuple of torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various elements depending on the configuration (ASTConfig) and inputs.

loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Classification (or regression if config.num_labels==1) loss.
logits (torch.FloatTensor of shape (batch_size, config.num_labels)) — Classification (or regression if config.num_labels==1) scores (before SoftMax).
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.FloatTensor (one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

The ASTForAudioClassification forward method, overrides the __call__ special method.

Example:

>>> from transformers import AutoFeatureExtractor, ASTForAudioClassification
>>> from datasets import load_dataset
>>> import torch

>>> dataset = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation", trust_remote_code=True)
>>> dataset = dataset.sort("id")
>>> sampling_rate = dataset.features["audio"].sampling_rate

>>> feature_extractor = AutoFeatureExtractor.from_pretrained("MIT/ast-finetuned-audioset-10-10-0.4593")
>>> model = ASTForAudioClassification.from_pretrained("MIT/ast-finetuned-audioset-10-10-0.4593")

>>> # audio file is decoded on the fly
>>> inputs = feature_extractor(dataset[0]["audio"]["array"], sampling_rate=sampling_rate, return_tensors="pt")

>>> with torch.no_grad():
...     logits = model(**inputs).logits

>>> predicted_class_ids = torch.argmax(logits, dim=-1).item()
>>> predicted_label = model.config.id2label[predicted_class_ids]
>>> predicted_label
'Speech'

>>> # compute loss - target_label is e.g. "down"
>>> target_label = model.config.id2label[0]
>>> inputs["labels"] = torch.tensor([model.config.label2id[target_label]])
>>> loss = model(**inputs).loss
>>> round(loss.item(), 2)
0.17

< > Update on GitHub

←ZoeDepth Bark→