SEW (Squeezed and Efficient Wav2Vec) was proposed in Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
The abstract from the paper is the following:
This paper is a study of performance-efficiency trade-offs in pre-trained models for automatic speech recognition (ASR). We focus on wav2vec 2.0, and formalize several architecture designs that influence both the model performance and its efficiency. Putting together all our observations, we introduce SEW (Squeezed and Efficient Wav2vec), a pre-trained model architecture with significant improvements along both performance and efficiency dimensions across a variety of training setups. For example, under the 100h-960h semi-supervised setup on LibriSpeech, SEW achieves a 1.9x inference speedup compared to wav2vec 2.0, with a 13.5% relative reduction in word error rate. With a similar inference time, SEW reduces word error rate by 25-50% across different model sizes.
This model was contributed by anton-l.
( vocab_size = 32 hidden_size = 768 num_hidden_layers = 12 num_attention_heads = 12 intermediate_size = 3072 squeeze_factor = 2 hidden_act = 'gelu' hidden_dropout = 0.1 activation_dropout = 0.1 attention_dropout = 0.1 feat_proj_dropout = 0.0 final_dropout = 0.1 layerdrop = 0.1 initializer_range = 0.02 layer_norm_eps = 1e-05 feat_extract_norm = 'group' feat_extract_activation = 'gelu' conv_dim = (64, 128, 128, 128, 128, 256, 256, 256, 256, 512, 512, 512, 512) conv_stride = (5, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1) conv_kernel = (10, 3, 1, 3, 1, 3, 1, 3, 1, 2, 1, 2, 1) conv_bias = False num_conv_pos_embeddings = 128 num_conv_pos_embedding_groups = 16 apply_spec_augment = True mask_time_prob = 0.05 mask_time_length = 10 mask_time_min_masks = 2 mask_feature_prob = 0.0 mask_feature_length = 10 mask_feature_min_masks = 0 ctc_loss_reduction = 'mean' ctc_zero_infinity = False use_weighted_layer_sum = False classifier_proj_size = 256 pad_token_id = 0 bos_token_id = 1 eos_token_id = 2 **kwargs )
Parameters
int, optional, defaults to 32) —
Vocabulary size of the SEW model. Defines the number of different tokens that can be represented by the
inputs_ids passed when calling SEW. int, optional, defaults to 768) —
Dimensionality of the encoder layers and the pooler layer. int, optional, defaults to 12) —
Number of hidden layers in the Transformer encoder. int, optional, defaults to 12) —
Number of attention heads for each attention layer in the Transformer encoder. int, optional, defaults to 3072) —
Dimensionality of the “intermediate” (i.e., feed-forward) layer in the Transformer encoder. int, optional, defaults to 2) —
Sequence length downsampling factor after the encoder and upsampling factor after the transformer. 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. float, optional, defaults to 0.1) —
The dropout probability for all fully connected layers in the embeddings, encoder, and pooler. float, optional, defaults to 0.1) —
The dropout ratio for activations inside the fully connected layer. float, optional, defaults to 0.1) —
The dropout ratio for the attention probabilities. float, optional, defaults to 0.1) —
The dropout probability for the final projection layer of SEWForCTC. float, optional, defaults to 0.1) —
The LayerDrop probability. See the [LayerDrop paper](see https://arxiv.org/abs/1909.11556) for more
details. float, optional, defaults to 0.02) —
The standard deviation of the truncated_normal_initializer for initializing all weight matrices. float, optional, defaults to 1e-12) —
The epsilon used by the layer normalization layers. str, optional, defaults to "group") —
The norm to be applied to 1D convolutional layers in feature encoder. One of "group" for group
normalization of only the first 1D convolutional layer or "layer" for layer normalization of all 1D
convolutional layers. float, optional, defaults to 0.0) —
The dropout probability for output of the feature encoder. str, optional, defaults to “gelu”) -- The non-linear activation function (function or string) in the 1D convolutional layers of the feature extractor. If string, “gelu”, “relu”, “selu”and“gelu_new”` are supported. Tuple[int] or List[int], optional, defaults to (64, 128, 128, 128, 128, 256, 256, 256, 256, 512, 512, 512, 512)) —
A tuple of integers defining the number of input and output channels of each 1D convolutional layer in the
feature encoder. The length of conv_dim defines the number of 1D convolutional layers. Tuple[int] or List[int], optional, defaults to (5, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1)) —
A tuple of integers defining the stride of each 1D convolutional layer in the feature encoder. The length
of conv_stride defines the number of convolutional layers and has to match the length of conv_dim. Tuple[int] or List[int], optional, defaults to (10, 3, 1, 3, 1, 3, 1, 3, 1, 2, 1, 2, 1)) —
A tuple of integers defining the kernel size of each 1D convolutional layer in the feature encoder. The
length of conv_kernel defines the number of convolutional layers and has to match the length of
conv_dim. bool, optional, defaults to False) —
Whether the 1D convolutional layers have a bias. int, optional, defaults to 128) —
Number of convolutional positional embeddings. Defines the kernel size of 1D convolutional positional
embeddings layer. int, optional, defaults to 16) —
Number of groups of 1D convolutional positional embeddings layer. bool, optional, defaults to True) —
Whether to apply SpecAugment data augmentation to the outputs of the feature encoder. For reference see
SpecAugment: A Simple Data Augmentation Method for Automatic Speech
Recognition. float, optional, defaults to 0.05) —
Percentage (between 0 and 1) of all feature vectors along the time axis which will be masked. The masking
procecure generates ”mask_time_problen(time_axis)/mask_time_length” independent masks over the axis. If
reasoning from the propability of each feature vector to be chosen as the start of the vector span to be
masked, mask_time_prob should be `prob_vector_startmask_time_length. Note that overlap may decrease the actual percentage of masked vectors. This is only relevant if apply_spec_augment is True`. int, optional, defaults to 10) —
Length of vector span along the time axis. int, optional, defaults to 2), —
The minimum number of masks of length mask_feature_length generated along the time axis, each time step,
irrespectively of mask_feature_prob. Only relevant if ”mask_time_prob*len(time_axis)/mask_time_length <
mask_time_min_masks” float, optional, defaults to 0.0) —
Percentage (between 0 and 1) of all feature vectors along the feature axis which will be masked. The
masking procecure generates ”mask_feature_problen(feature_axis)/mask_time_length” independent masks over
the axis. If reasoning from the propability of each feature vector to be chosen as the start of the vector
span to be masked, mask_feature_prob should be `prob_vector_startmask_feature_length. Note that overlap may decrease the actual percentage of masked vectors. This is only relevant if apply_spec_augment is
True`. int, optional, defaults to 10) —
Length of vector span along the feature axis. int, optional, defaults to 0), —
The minimum number of masks of length mask_feature_length generated along the feature axis, each time
step, irrespectively of mask_feature_prob. Only relevant if
”mask_feature_prob*len(feature_axis)/mask_feature_length < mask_feature_min_masks” str, optional, defaults to "sum") —
Specifies the reduction to apply to the output of torch.nn.CTCLoss. Only relevant when training an
instance of SEWForCTC. bool, optional, defaults to False) —
Whether to zero infinite losses and the associated gradients of torch.nn.CTCLoss. Infinite losses mainly
occur when the inputs are too short to be aligned to the targets. Only relevant when training an instance
of SEWForCTC. bool, optional, defaults to False) —
Whether to use a weighted average of layer outputs with learned weights. Only relevant when using an
instance of Wav2Vec2ForSequenceClassification. int, optional, defaults to 256) —
Dimensionality of the projection before token mean-pooling for classification. This is the configuration class to store the configuration of a SEWModel. It is used to instantiate a SEW 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 SEW asapp/sew-tiny-100k 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 SEWConfig, SEWModel
>>> # Initializing a SEW asapp/sew-tiny-100k style configuration
>>> configuration = SEWConfig()
>>> # Initializing a model (with random weights) from the asapp/sew-tiny-100k style configuration
>>> model = SEWModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config( config: SEWConfig )
Parameters
The bare SEW Model transformer outputting raw hidden-states without any specific head on top. SEW was proposed in Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving etc.).
This model is a PyTorch torch.nn.Module sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_values: Optional attention_mask: Optional = None mask_time_indices: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None ) → transformers.modeling_outputs.BaseModelOutput or tuple(torch.FloatTensor)
Parameters
torch.FloatTensor of shape (batch_size, sequence_length)) —
Float values of input raw speech waveform. Values 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_values, the AutoProcessor should be used for padding and
conversion into a tensor of type torch.FloatTensor. See Wav2Vec2Processor.call() for details. torch.LongTensor of shape (batch_size, sequence_length), optional) —
Mask to avoid performing convolution and attention on padding token indices. Mask values selected in [0, 1]:
bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. Returns
transformers.modeling_outputs.BaseModelOutput or tuple(torch.FloatTensor)
A transformers.modeling_outputs.BaseModelOutput 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 (SEWConfig) 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.
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 SEWModel 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, SEWModel
>>> import torch
>>> from datasets import load_dataset
>>> dataset = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation")
>>> dataset = dataset.sort("id")
>>> sampling_rate = dataset.features["audio"].sampling_rate
>>> processor = AutoProcessor.from_pretrained("asapp/sew-tiny-100k-ft-ls100h")
>>> model = SEWModel.from_pretrained("asapp/sew-tiny-100k-ft-ls100h")
>>> # 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, 292, 512]( config target_lang: Optional = None )
Parameters
SEW Model with a language modeling head on top for Connectionist Temporal Classification (CTC).
SEW was proposed in Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech
Recognition by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger,
Yoav Artzi.
This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving etc.).
This model is a PyTorch torch.nn.Module sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_values: Optional attention_mask: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None labels: Optional = None ) → transformers.modeling_outputs.CausalLMOutput or tuple(torch.FloatTensor)
Parameters
torch.FloatTensor of shape (batch_size, sequence_length)) —
Float values of input raw speech waveform. Values 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_values, the AutoProcessor should be used for padding and
conversion into a tensor of type torch.FloatTensor. See Wav2Vec2Processor.call() for details. torch.LongTensor of shape (batch_size, sequence_length), optional) —
Mask to avoid performing convolution and attention on padding token indices. Mask values selected in [0, 1]:
bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. torch.LongTensor of shape (batch_size, target_length), optional) —
Labels for connectionist temporal classification. Note that target_length has to be smaller or equal to
the sequence length of the output logits. Indices are selected in [-100, 0, ..., config.vocab_size - 1].
All labels set to -100 are ignored (masked), the loss is only computed for labels in [0, ..., config.vocab_size - 1]. Returns
transformers.modeling_outputs.CausalLMOutput or tuple(torch.FloatTensor)
A transformers.modeling_outputs.CausalLMOutput 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 (SEWConfig) and inputs.
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Language modeling loss (for next-token prediction).
logits (torch.FloatTensor of shape (batch_size, sequence_length, config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token 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 SEWForCTC 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, SEWForCTC
>>> from datasets import load_dataset
>>> import torch
>>> dataset = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation")
>>> dataset = dataset.sort("id")
>>> sampling_rate = dataset.features["audio"].sampling_rate
>>> processor = AutoProcessor.from_pretrained("asapp/sew-tiny-100k-ft-ls100h")
>>> model = SEWForCTC.from_pretrained("asapp/sew-tiny-100k-ft-ls100h")
>>> # audio file is decoded on the fly
>>> inputs = processor(dataset[0]["audio"]["array"], sampling_rate=sampling_rate, return_tensors="pt")
>>> with torch.no_grad():
... logits = model(**inputs).logits
>>> predicted_ids = torch.argmax(logits, dim=-1)
>>> # transcribe speech
>>> transcription = processor.batch_decode(predicted_ids)
>>> transcription[0]
'MISTER QUILTER IS THE APPOSTILE OF THE MIDDLE CLASSES AND WE ARE GLAD TO WELCOME HIS GOSPOLLE'
>>> inputs["labels"] = processor(text=dataset[0]["text"], return_tensors="pt").input_ids
>>> # compute loss
>>> loss = model(**inputs).loss
>>> round(loss.item(), 2)
0.42( config )
Parameters
SEW Model with a sequence classification head on top (a linear layer over the pooled output) for tasks like SUPERB Keyword Spotting.
SEW was proposed in Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving etc.).
This model is a PyTorch torch.nn.Module sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_values: Optional attention_mask: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None labels: Optional = None ) → transformers.modeling_outputs.SequenceClassifierOutput or tuple(torch.FloatTensor)
Parameters
torch.FloatTensor of shape (batch_size, sequence_length)) —
Float values of input raw speech waveform. Values 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_values, the AutoProcessor should be used for padding and
conversion into a tensor of type torch.FloatTensor. See Wav2Vec2Processor.call() for details. torch.LongTensor of shape (batch_size, sequence_length), optional) —
Mask to avoid performing convolution and attention on padding token indices. Mask values selected in [0, 1]:
bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. torch.LongTensor of shape (batch_size,), optional) —
Labels for computing the sequence 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 (SEWConfig) 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 SEWForSequenceClassification 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 AutoFeatureExtractor, SEWForSequenceClassification
>>> from datasets import load_dataset
>>> import torch
>>> dataset = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation")
>>> dataset = dataset.sort("id")
>>> sampling_rate = dataset.features["audio"].sampling_rate
>>> feature_extractor = AutoFeatureExtractor.from_pretrained("anton-l/sew-mid-100k-ft-keyword-spotting")
>>> model = SEWForSequenceClassification.from_pretrained("anton-l/sew-mid-100k-ft-keyword-spotting")
>>> # 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
'_unknown_'
>>> # 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)
9.52