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.`;return{c(){t=p("p"),t.innerHTML=T},l(l){t=m(l,"P",{"data-svelte-h":!0}),v(t)!=="svelte-fincs2"&&(t.innerHTML=T)},m(l,d){c(l,t,d)},p:Y,d(l){l&&s(t)}}}function Go(U){let t,T="Example:",l,d,y;return d=new Oe({props:{code:"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",highlighted:`from transformers import AutoProcessor, UniSpeechModel
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("patrickvonplaten/unispeech-large-1500h-cv-timit")
model = UniSpeechModel.from_pretrained("patrickvonplaten/unispeech-large-1500h-cv-timit")
# 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, 1024]`,wrap:!1}}),{c(){t=p("p"),t.textContent=T,l=a(),h(d.$$.fragment)},l(o){t=m(o,"P",{"data-svelte-h":!0}),v(t)!=="svelte-11lpom8"&&(t.textContent=T),l=r(o),u(d.$$.fragment,o)},m(o,w){c(o,t,w),c(o,l,w),f(d,o,w),y=!0},p:Y,i(o){y||(g(d.$$.fragment,o),y=!0)},o(o){_(d.$$.fragment,o),y=!1},d(o){o&&(s(t),s(l)),b(d,o)}}}function qo(U){let t,T=`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.`;return{c(){t=p("p"),t.innerHTML=T},l(l){t=m(l,"P",{"data-svelte-h":!0}),v(t)!=="svelte-fincs2"&&(t.innerHTML=T)},m(l,d){c(l,t,d)},p:Y,d(l){l&&s(t)}}}function Vo(U){let t,T="Example:",l,d,y;return d=new Oe({props:{code:"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",highlighted:`from transformers import AutoProcessor, UniSpeechForCTC
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
processor = AutoProcessor.from_pretrained("patrickvonplaten/unispeech-large-1500h-cv-timit")
model = UniSpeechForCTC.from_pretrained("patrickvonplaten/unispeech-large-1500h-cv-timit")
# 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 apposl of the midle classes and weare glad to welcom his gosepl'
inputs["labels"] = processor(text=dataset[0]["text"], return_tensors="pt").input_ids
# compute loss
loss = model(**inputs).loss
round(loss.item(), 2)
17.17`,wrap:!1}}),{c(){t=p("p"),t.textContent=T,l=a(),h(d.$$.fragment)},l(o){t=m(o,"P",{"data-svelte-h":!0}),v(t)!=="svelte-11lpom8"&&(t.textContent=T),l=r(o),u(d.$$.fragment,o)},m(o,w){c(o,t,w),c(o,l,w),f(d,o,w),y=!0},p:Y,i(o){y||(g(d.$$.fragment,o),y=!0)},o(o){_(d.$$.fragment,o),y=!1},d(o){o&&(s(t),s(l)),b(d,o)}}}function No(U){let t,T=`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.`;return{c(){t=p("p"),t.innerHTML=T},l(l){t=m(l,"P",{"data-svelte-h":!0}),v(t)!=="svelte-fincs2"&&(t.innerHTML=T)},m(l,d){c(l,t,d)},p:Y,d(l){l&&s(t)}}}function Xo(U){let t,T="Example:",l,d,y;return d=new Oe({props:{code:"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",highlighted:`from transformers import AutoFeatureExtractor, UniSpeechForSequenceClassification
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("patrickvonplaten/unispeech-large-1500h-cv-timit")
model = UniSpeechForSequenceClassification.from_pretrained("patrickvonplaten/unispeech-large-1500h-cv-timit")
# 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]
# 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`,wrap:!1}}),{c(){t=p("p"),t.textContent=T,l=a(),h(d.$$.fragment)},l(o){t=m(o,"P",{"data-svelte-h":!0}),v(t)!=="svelte-11lpom8"&&(t.textContent=T),l=r(o),u(d.$$.fragment,o)},m(o,w){c(o,t,w),c(o,l,w),f(d,o,w),y=!0},p:Y,i(o){y||(g(d.$$.fragment,o),y=!0)},o(o){_(d.$$.fragment,o),y=!1},d(o){o&&(s(t),s(l)),b(d,o)}}}function Ro(U){let t,T=`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.`;return{c(){t=p("p"),t.innerHTML=T},l(l){t=m(l,"P",{"data-svelte-h":!0}),v(t)!=="svelte-fincs2"&&(t.innerHTML=T)},m(l,d){c(l,t,d)},p:Y,d(l){l&&s(t)}}}function Io(U){let t,T="Example:",l,d,y;return d=new Oe({props:{code:"aW1wb3J0JTIwdG9yY2glMEFmcm9tJTIwdHJhbnNmb3JtZXJzJTIwaW1wb3J0JTIwQXV0b0ZlYXR1cmVFeHRyYWN0b3IlMkMlMjBVbmlTcGVlY2hGb3JQcmVUcmFpbmluZyUwQSUwQWZlYXR1cmVfZXh0cmFjdG9yJTIwJTNEJTIwQXV0b0ZlYXR1cmVFeHRyYWN0b3IuZnJvbV9wcmV0cmFpbmVkKCUyMm1pY3Jvc29mdCUyRnVuaXNwZWVjaC1sYXJnZS0xNTAwaC1jdiUyMiklMEFtb2RlbCUyMCUzRCUyMFVuaVNwZWVjaEZvclByZVRyYWluaW5nLmZyb21fcHJldHJhaW5lZCglMjJtaWNyb3NvZnQlMkZ1bmlzcGVlY2gtbGFyZ2UtMTUwMGgtY3YlMjIpJTBBJTIzJTIwVE9ETyUzQSUyMEFkZCUyMGZ1bGwlMjBwcmV0cmFpbmluZyUyMGV4YW1wbGU=",highlighted:`import torch
from transformers import AutoFeatureExtractor, UniSpeechForPreTraining
feature_extractor = AutoFeatureExtractor.from_pretrained("microsoft/unispeech-large-1500h-cv")
model = UniSpeechForPreTraining.from_pretrained("microsoft/unispeech-large-1500h-cv")
# TODO: Add full pretraining example`,wrap:!1}}),{c(){t=p("p"),t.textContent=T,l=a(),h(d.$$.fragment)},l(o){t=m(o,"P",{"data-svelte-h":!0}),v(t)!=="svelte-11lpom8"&&(t.textContent=T),l=r(o),u(d.$$.fragment,o)},m(o,w){c(o,t,w),c(o,l,w),f(d,o,w),y=!0},p:Y,i(o){y||(g(d.$$.fragment,o),y=!0)},o(o){_(d.$$.fragment,o),y=!1},d(o){o&&(s(t),s(l)),b(d,o)}}}function Lo(U){let t,T,l,d,y,o,w,Ae,ne,to=`The UniSpeech model was proposed in UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael
Zeng, Xuedong Huang .`,Ke,se,oo="The abstract from the paper is the following:",et,ae,no=`In this paper, we propose a unified pre-training approach called UniSpeech to learn speech representations with both
unlabeled and labeled data, in which supervised phonetic CTC learning and phonetically-aware contrastive
self-supervised learning are conducted in a multi-task learning manner. The resultant representations can capture
information more correlated with phonetic structures and improve the generalization across languages and domains. We
evaluate the effectiveness of UniSpeech for cross-lingual representation learning on public CommonVoice corpus. The
results show that UniSpeech outperforms self-supervised pretraining and supervised transfer learning for speech
recognition by a maximum of 13.4% and 17.8% relative phone error rate reductions respectively (averaged over all
testing languages). The transferability of UniSpeech is also demonstrated on a domain-shift speech recognition task,
i.e., a relative word error rate reduction of 6% against the previous approach.`,tt,re,so=`This model was contributed by patrickvonplaten. The Authors’ code can be
found here.`,ot,ie,nt,le,ao=`UniSpeechForPreTrainingOutput, with potential hidden states and attentions.",dt,fe,pt,C,ge,kt,Je,po=`The bare UniSpeech Model transformer outputting raw hidden-states without any specific head on top.
UniSpeech was proposed in UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled
Data by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei,
Michael Zeng, Xuedong Huang.`,$t,We,mo=`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.).`,St,Ze,ho=`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.`,jt,W,_e,xt,ze,uo='The UniSpeechModel forward method, overrides the __call__ special method.',Ft,Q,Jt,E,mt,be,ht,k,ve,Wt,Ge,fo=`UniSpeech Model with a language modeling head on top for Connectionist Temporal Classification (CTC).
UniSpeech was proposed in UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled
Data by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei,
Michael Zeng, Xuedong Huang.`,Zt,qe,go=`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.).`,zt,Ve,_o=`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.`,Gt,Z,ye,qt,Ne,bo='The UniSpeechForCTC forward method, overrides the __call__ special method.',Vt,D,Nt,O,ut,Te,ft,M,we,Xt,Xe,vo=`UniSpeech Model with a sequence classification head on top (a linear layer over the pooled output) for tasks like
SUPERB Keyword Spotting.`,Rt,Re,yo=`UniSpeech was proposed in UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled
Data by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei,
Michael Zeng, Xuedong Huang.`,It,Ie,To=`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.).`,Lt,Le,wo=`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.`,Bt,z,Ue,Ht,Be,Uo='The UniSpeechForSequenceClassification forward method, overrides the __call__ special method.',Yt,A,Pt,K,gt,Me,_t,$,Ce,Qt,He,Mo=`UniSpeech Model with a vector-quantization module and ctc loss for pre-training.
UniSpeech was proposed in UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled
Data by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei,
Michael Zeng, Xuedong Huang.`,Et,Ye,Co=`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.).`,Dt,Pe,ko=`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.`,Ot,G,ke,At,Qe,$o='The UniSpeechForPreTraining forward method, overrides the __call__ special method.',Kt,ee,eo,te,bt,$e,vt,Ee,yt;return y=new B({props:{title:"UniSpeech",local:"unispeech",headingTag:"h1"}}),w=new B({props:{title:"Overview",local:"overview",headingTag:"h2"}}),ie=new B({props:{title:"Usage tips",local:"usage-tips",headingTag:"h2"}}),ce=new B({props:{title:"Resources",local:"resources",headingTag:"h2"}}),pe=new B({props:{title:"UniSpeechConfig",local:"transformers.UniSpeechConfig",headingTag:"h2"}}),me=new L({props:{name:"class transformers.UniSpeechConfig",anchor:"transformers.UniSpeechConfig",parameters:[{name:"vocab_size",val:" = 32"},{name:"hidden_size",val:" = 768"},{name:"num_hidden_layers",val:" = 12"},{name:"num_attention_heads",val:" = 12"},{name:"intermediate_size",val:" = 3072"},{name:"hidden_act",val:" = 'gelu'"},{name:"hidden_dropout",val:" = 0.1"},{name:"activation_dropout",val:" = 0.1"},{name:"attention_dropout",val:" = 0.1"},{name:"feat_proj_dropout",val:" = 0.0"},{name:"feat_quantizer_dropout",val:" = 0.0"},{name:"final_dropout",val:" = 0.1"},{name:"layerdrop",val:" = 0.1"},{name:"initializer_range",val:" = 0.02"},{name:"layer_norm_eps",val:" = 1e-05"},{name:"feat_extract_norm",val:" = 'group'"},{name:"feat_extract_activation",val:" = 'gelu'"},{name:"conv_dim",val:" = (512, 512, 512, 512, 512, 512, 512)"},{name:"conv_stride",val:" = (5, 2, 2, 2, 2, 2, 2)"},{name:"conv_kernel",val:" = (10, 3, 3, 3, 3, 2, 2)"},{name:"conv_bias",val:" = False"},{name:"num_conv_pos_embeddings",val:" = 128"},{name:"num_conv_pos_embedding_groups",val:" = 16"},{name:"do_stable_layer_norm",val:" = False"},{name:"apply_spec_augment",val:" = True"},{name:"mask_time_prob",val:" = 0.05"},{name:"mask_time_length",val:" = 10"},{name:"mask_time_min_masks",val:" = 2"},{name:"mask_feature_prob",val:" = 0.0"},{name:"mask_feature_length",val:" = 10"},{name:"mask_feature_min_masks",val:" = 0"},{name:"num_codevectors_per_group",val:" = 320"},{name:"num_codevector_groups",val:" = 2"},{name:"contrastive_logits_temperature",val:" = 0.1"},{name:"num_negatives",val:" = 100"},{name:"codevector_dim",val:" = 256"},{name:"proj_codevector_dim",val:" = 256"},{name:"diversity_loss_weight",val:" = 0.1"},{name:"ctc_loss_reduction",val:" = 'mean'"},{name:"ctc_zero_infinity",val:" = False"},{name:"use_weighted_layer_sum",val:" = False"},{name:"classifier_proj_size",val:" = 256"},{name:"num_ctc_classes",val:" = 80"},{name:"pad_token_id",val:" = 0"},{name:"bos_token_id",val:" = 1"},{name:"eos_token_id",val:" = 2"},{name:"replace_prob",val:" = 0.5"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"transformers.UniSpeechConfig.vocab_size",description:`vocab_size (int, optional, defaults to 32) —
Vocabulary size of the UniSpeech model. Defines the number of different tokens that can be represented by
the inputs_ids passed when calling UniSpeechModel. Vocabulary size of the model. Defines the
different tokens that can be represented by the inputs_ids passed to the forward method of
UniSpeechModel.`,name:"vocab_size"},{anchor:"transformers.UniSpeechConfig.hidden_size",description:`hidden_size (int, optional, defaults to 768) —
Dimensionality of the encoder layers and the pooler layer.`,name:"hidden_size"},{anchor:"transformers.UniSpeechConfig.num_hidden_layers",description:`num_hidden_layers (int, optional, defaults to 12) —
Number of hidden layers in the Transformer encoder.`,name:"num_hidden_layers"},{anchor:"transformers.UniSpeechConfig.num_attention_heads",description:`num_attention_heads (int, optional, defaults to 12) —
Number of attention heads for each attention layer in the Transformer encoder.`,name:"num_attention_heads"},{anchor:"transformers.UniSpeechConfig.intermediate_size",description:`intermediate_size (int, optional, defaults to 3072) —
Dimensionality of the “intermediate” (i.e., feed-forward) layer in the Transformer encoder.`,name:"intermediate_size"},{anchor:"transformers.UniSpeechConfig.hidden_act",description:`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.`,name:"hidden_act"},{anchor:"transformers.UniSpeechConfig.hidden_dropout",description:`hidden_dropout (float, optional, defaults to 0.1) —
The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.`,name:"hidden_dropout"},{anchor:"transformers.UniSpeechConfig.activation_dropout",description:`activation_dropout (float, optional, defaults to 0.1) —
The dropout ratio for activations inside the fully connected layer.`,name:"activation_dropout"},{anchor:"transformers.UniSpeechConfig.attention_dropout",description:`attention_dropout (float, optional, defaults to 0.1) —
The dropout ratio for the attention probabilities.`,name:"attention_dropout"},{anchor:"transformers.UniSpeechConfig.feat_proj_dropout",description:`feat_proj_dropout (float, optional, defaults to 0.0) —
The dropout probability for output of the feature encoder.`,name:"feat_proj_dropout"},{anchor:"transformers.UniSpeechConfig.feat_quantizer_dropout",description:`feat_quantizer_dropout (float, optional, defaults to 0.0) —
The dropout probability for the output of the feature encoder that’s used by the quantizer.`,name:"feat_quantizer_dropout"},{anchor:"transformers.UniSpeechConfig.final_dropout",description:`final_dropout (float, optional, defaults to 0.1) —
The dropout probability for the final projection layer of UniSpeechForCTC.`,name:"final_dropout"},{anchor:"transformers.UniSpeechConfig.layerdrop",description:`layerdrop (float, optional, defaults to 0.1) —
The LayerDrop probability. See the [LayerDrop paper](see https://arxiv.org/abs/1909.11556) for more
details.`,name:"layerdrop"},{anchor:"transformers.UniSpeechConfig.initializer_range",description:`initializer_range (float, optional, defaults to 0.02) —
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.`,name:"initializer_range"},{anchor:"transformers.UniSpeechConfig.layer_norm_eps",description:`layer_norm_eps (float, optional, defaults to 1e-05) —
The epsilon used by the layer normalization layers.`,name:"layer_norm_eps"},{anchor:"transformers.UniSpeechConfig.feat_extract_norm",description:`feat_extract_norm (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.`,name:"feat_extract_norm"},{anchor:"transformers.UniSpeechConfig.feat_extract_activation",description:"feat_extract_activation (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.",name:"feat_extract_activation"},{anchor:"transformers.UniSpeechConfig.conv_dim",description:`conv_dim (Tuple[int] or List[int], optional, defaults to (512, 512, 512, 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.`,name:"conv_dim"},{anchor:"transformers.UniSpeechConfig.conv_stride",description:`conv_stride (Tuple[int] or List[int], optional, defaults to (5, 2, 2, 2, 2, 2, 2)) —
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.`,name:"conv_stride"},{anchor:"transformers.UniSpeechConfig.conv_kernel",description:`conv_kernel (Tuple[int] or List[int], optional, defaults to (10, 3, 3, 3, 3, 2, 2)) —
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.`,name:"conv_kernel"},{anchor:"transformers.UniSpeechConfig.conv_bias",description:`conv_bias (bool, optional, defaults to False) —
Whether the 1D convolutional layers have a bias.`,name:"conv_bias"},{anchor:"transformers.UniSpeechConfig.num_conv_pos_embeddings",description:`num_conv_pos_embeddings (int, optional, defaults to 128) —
Number of convolutional positional embeddings. Defines the kernel size of 1D convolutional positional
embeddings layer.`,name:"num_conv_pos_embeddings"},{anchor:"transformers.UniSpeechConfig.num_conv_pos_embedding_groups",description:`num_conv_pos_embedding_groups (int, optional, defaults to 16) —
Number of groups of 1D convolutional positional embeddings layer.`,name:"num_conv_pos_embedding_groups"},{anchor:"transformers.UniSpeechConfig.do_stable_layer_norm",description:`do_stable_layer_norm (bool, optional, defaults to False) —
Whether to apply stable layer norm architecture of the Transformer encoder. do_stable_layer_norm is True corresponds to applying layer norm before the attention layer, whereas do_stable_layer_norm is False corresponds to applying layer norm after the attention layer.`,name:"do_stable_layer_norm"},{anchor:"transformers.UniSpeechConfig.apply_spec_augment",description:`apply_spec_augment (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.`,name:"apply_spec_augment"},{anchor:"transformers.UniSpeechConfig.mask_time_prob",description:`mask_time_prob (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\`.`,name:"mask_time_prob"},{anchor:"transformers.UniSpeechConfig.mask_time_length",description:`mask_time_length (int, optional, defaults to 10) —
Length of vector span along the time axis.`,name:"mask_time_length"},{anchor:"transformers.UniSpeechConfig.mask_time_min_masks",description:`mask_time_min_masks (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”`,name:"mask_time_min_masks"},{anchor:"transformers.UniSpeechConfig.mask_feature_prob",description:`mask_feature_prob (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\`.`,name:"mask_feature_prob"},{anchor:"transformers.UniSpeechConfig.mask_feature_length",description:`mask_feature_length (int, optional, defaults to 10) —
Length of vector span along the feature axis.`,name:"mask_feature_length"},{anchor:"transformers.UniSpeechConfig.mask_feature_min_masks",description:`mask_feature_min_masks (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”`,name:"mask_feature_min_masks"},{anchor:"transformers.UniSpeechConfig.num_codevectors_per_group",description:`num_codevectors_per_group (int, optional, defaults to 320) —
Number of entries in each quantization codebook (group).`,name:"num_codevectors_per_group"},{anchor:"transformers.UniSpeechConfig.num_codevector_groups",description:`num_codevector_groups (int, optional, defaults to 2) —
Number of codevector groups for product codevector quantization.`,name:"num_codevector_groups"},{anchor:"transformers.UniSpeechConfig.contrastive_logits_temperature",description:`contrastive_logits_temperature (float, optional, defaults to 0.1) —
The temperature kappa in the contrastive loss.`,name:"contrastive_logits_temperature"},{anchor:"transformers.UniSpeechConfig.num_negatives",description:`num_negatives (int, optional, defaults to 100) —
Number of negative samples for the contrastive loss.`,name:"num_negatives"},{anchor:"transformers.UniSpeechConfig.codevector_dim",description:`codevector_dim (int, optional, defaults to 256) —
Dimensionality of the quantized feature vectors.`,name:"codevector_dim"},{anchor:"transformers.UniSpeechConfig.proj_codevector_dim",description:`proj_codevector_dim (int, optional, defaults to 256) —
Dimensionality of the final projection of both the quantized and the transformer features.`,name:"proj_codevector_dim"},{anchor:"transformers.UniSpeechConfig.diversity_loss_weight",description:`diversity_loss_weight (int, optional, defaults to 0.1) —
The weight of the codebook diversity loss component.`,name:"diversity_loss_weight"},{anchor:"transformers.UniSpeechConfig.ctc_loss_reduction",description:`ctc_loss_reduction (str, optional, defaults to "mean") —
Specifies the reduction to apply to the output of torch.nn.CTCLoss. Only relevant when training an
instance of UniSpeechForCTC.`,name:"ctc_loss_reduction"},{anchor:"transformers.UniSpeechConfig.ctc_zero_infinity",description:`ctc_zero_infinity (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 UniSpeechForCTC.`,name:"ctc_zero_infinity"},{anchor:"transformers.UniSpeechConfig.use_weighted_layer_sum",description:`use_weighted_layer_sum (bool, optional, defaults to False) —
Whether to use a weighted average of layer outputs with learned weights. Only relevant when using an
instance of UniSpeechForSequenceClassification.`,name:"use_weighted_layer_sum"},{anchor:"transformers.UniSpeechConfig.classifier_proj_size",description:`classifier_proj_size (int, optional, defaults to 256) —
Dimensionality of the projection before token mean-pooling for classification.`,name:"classifier_proj_size"},{anchor:"transformers.UniSpeechConfig.num_ctc_classes",description:`num_ctc_classes (int, optional, defaults to 80) —
Specifies the number of classes (phoneme tokens and blank token) for phoneme-level CTC loss. Only relevant
when using an instance of UniSpeechForPreTraining.`,name:"num_ctc_classes"},{anchor:"transformers.UniSpeechConfig.pad_token_id",description:`pad_token_id (int, optional, defaults to 0) —
The id of the padding token.`,name:"pad_token_id"},{anchor:"transformers.UniSpeechConfig.bos_token_id",description:`bos_token_id (int, optional, defaults to 1) —
The id of the “beginning-of-sequence” token.`,name:"bos_token_id"},{anchor:"transformers.UniSpeechConfig.eos_token_id",description:`eos_token_id (int, optional, defaults to 2) —
The id of the “end-of-sequence” token.`,name:"eos_token_id"},{anchor:"transformers.UniSpeechConfig.replace_prob",description:`replace_prob (float, optional, defaults to 0.5) —
Propability that transformer feature is replaced by quantized feature for pretraining.`,name:"replace_prob"}],source:"https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/unispeech/configuration_unispeech.py#L27"}}),P=new De({props:{anchor:"transformers.UniSpeechConfig.example",$$slots:{default:[Zo]},$$scope:{ctx:U}}}),he=new B({props:{title:"UniSpeech specific outputs",local:"transformers.models.unispeech.modeling_unispeech.UniSpeechForPreTrainingOutput",headingTag:"h2"}}),ue=new L({props:{name:"class transformers.models.unispeech.modeling_unispeech.UniSpeechForPreTrainingOutput",anchor:"transformers.models.unispeech.modeling_unispeech.UniSpeechForPreTrainingOutput",parameters:[{name:"loss",val:": typing.Optional[torch.FloatTensor] = None"},{name:"projected_states",val:": FloatTensor = None"},{name:"projected_quantized_states",val:": FloatTensor = None"},{name:"codevector_perplexity",val:": FloatTensor = None"},{name:"hidden_states",val:": typing.Optional[typing.Tuple[torch.FloatTensor]] = None"},{name:"attentions",val:": typing.Optional[typing.Tuple[torch.FloatTensor]] = None"}],parametersDescription:[{anchor:"transformers.models.unispeech.modeling_unispeech.UniSpeechForPreTrainingOutput.loss",description:`loss (optional, returned when model is in train mode, torch.FloatTensor of shape (1,)) —
Total loss as the sum of the contrastive loss (L_m) and the diversity loss (L_d) as stated in the official
paper . (classification) loss.`,name:"loss"},{anchor:"transformers.models.unispeech.modeling_unispeech.UniSpeechForPreTrainingOutput.projected_states",description:`projected_states (torch.FloatTensor of shape (batch_size, sequence_length, config.proj_codevector_dim)) —
Hidden-states of the model projected to config.proj_codevector_dim that can be used to predict the masked
projected quantized states.`,name:"projected_states"},{anchor:"transformers.models.unispeech.modeling_unispeech.UniSpeechForPreTrainingOutput.projected_quantized_states",description:`projected_quantized_states (torch.FloatTensor of shape (batch_size, sequence_length, config.proj_codevector_dim)) —
Quantized extracted feature vectors projected to config.proj_codevector_dim representing the positive
target vectors for contrastive loss.`,name:"projected_quantized_states"},{anchor:"transformers.models.unispeech.modeling_unispeech.UniSpeechForPreTrainingOutput.hidden_states",description:`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 + 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 initial embedding outputs.`,name:"hidden_states"},{anchor:"transformers.models.unispeech.modeling_unispeech.UniSpeechForPreTrainingOutput.attentions",description:`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.`,name:"attentions"}],source:"https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/unispeech/modeling_unispeech.py#L67"}}),fe=new B({props:{title:"UniSpeechModel",local:"transformers.UniSpeechModel",headingTag:"h2"}}),ge=new L({props:{name:"class transformers.UniSpeechModel",anchor:"transformers.UniSpeechModel",parameters:[{name:"config",val:": UniSpeechConfig"}],parametersDescription:[{anchor:"transformers.UniSpeechModel.config",description:`config (UniSpeechConfig) — 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.`,name:"config"}],source:"https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/unispeech/modeling_unispeech.py#L1344"}}),_e=new L({props:{name:"forward",anchor:"transformers.UniSpeechModel.forward",parameters:[{name:"input_values",val:": typing.Optional[torch.Tensor]"},{name:"attention_mask",val:": typing.Optional[torch.Tensor] = None"},{name:"mask_time_indices",val:": typing.Optional[torch.FloatTensor] = None"},{name:"output_attentions",val:": typing.Optional[bool] = None"},{name:"output_hidden_states",val:": typing.Optional[bool] = None"},{name:"return_dict",val:": typing.Optional[bool] = None"}],parametersDescription:[{anchor:"transformers.UniSpeechModel.forward.input_values",description:`input_values (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.`,name:"input_values"},{anchor:"transformers.UniSpeechModel.forward.attention_mask",description:`attention_mask (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]:
attention_mask should only be passed if the corresponding processor has config.return_attention_mask == True. For all models whose processor has config.return_attention_mask == False, attention_mask should
not be passed to avoid degraded performance when doing batched inference. For such models
input_values should simply be padded with 0 and passed without attention_mask. Be aware that these
models also yield slightly different results depending on whether input_values is padded or not.
bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail.`,name:"output_attentions"},{anchor:"transformers.UniSpeechModel.forward.output_hidden_states",description:`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.`,name:"output_hidden_states"},{anchor:"transformers.UniSpeechModel.forward.return_dict",description:`return_dict (bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple.`,name:"return_dict"}],source:"https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/unispeech/modeling_unispeech.py#L1413",returnDescription:`