"""This lobe enables the integration of fairseq pretrained wav2vec models.
Reference: https://arxiv.org/abs/2006.11477
Reference: https://arxiv.org/abs/1904.05862
FairSeq >= 1.0.0 needs to be installed: https://fairseq.readthedocs.io/en/latest/
Authors
* Titouan Parcollet 2021
* Salima Mdhaffar 2021
"""
import torch
import logging
import torch.nn.functional as F
from torch import nn
from speechbrain.utils.data_utils import download_file
from speechbrain.dataio.dataio import length_to_mask
# We check if fairseq is installed.
try:
import fairseq
except ImportError:
MSG = "Please install Fairseq to use pretrained wav2vec\n"
MSG += "E.G. run: pip install fairseq"
raise ImportError(MSG)
logger = logging.getLogger(__name__)
[docs]
class FairseqWav2Vec2(nn.Module):
"""This lobe enables the integration of fairseq pretrained wav2vec2.0 models.
Source paper: https://arxiv.org/abs/2006.11477
FairSeq >= 0.10.0 needs to be installed:
https://fairseq.readthedocs.io/en/latest/
The model can be used as a fixed features extractor or can be finetuned. It
will download automatically the model if a url is given (e.g FairSeq
repository from GitHub).
Arguments
---------
pretrained_path : str
Path of the pretrained wav2vec2 model. It can be a url or a local path.
save_path : str
Path and filename of the downloaded model.
input_norm : bool (default: None)
If True, a layer_norm (affine) will be applied to the input waveform.
By default, it is extracted from the checkpoint of the downloaded model
in order to match the pretraining conditions. However, if this information
is not given in the checkpoint, it has to be given manually.
output_norm : bool (default: True)
If True, a layer_norm (affine) will be applied to the output obtained
from the wav2vec model.
freeze : bool (default: True)
If True, the model is frozen. If False, the model will be trained
alongside with the rest of the pipeline.
pretrain : bool (default: True)
If True, the model is pretrained with the specified source.
If False, the randomly-initialized model is instantiated.
dropout : float (default: None)
If different from None (0.0 to 1.0), it will override the given fairseq
dropout rates. This is useful if the wav2vec2 model has been trained
without dropout and one wants to reactivate it for downstream task
fine-tuning (better performance observed).
layer_drop : float (default: None)
If different from None (0.0 to 1.0), it will override the given fairseq
layer_drop rate. This is useful if the wav2vec2 model has been trained
without layer_drop and one wants to reactivate it for downstream task
fine-tuning.
Example
-------
>>> inputs = torch.rand([10, 600])
>>> model_url = "https://dl.fbaipublicfiles.com/fairseq/wav2vec/wav2vec_small.pt"
>>> save_path = "models_checkpoints/wav2vec2.pt"
>>> model = FairseqWav2Vec2(model_url, save_path)
>>> outputs = model(inputs)
>>> outputs.shape
torch.Size([10, 100, 768])
"""
def __init__(
self,
pretrained_path,
save_path,
input_norm=None,
output_norm=False,
freeze=False,
freeze_feature_extractor=False,
pretrain=True,
dropout=None,
layer_drop=None,
):
super().__init__()
# Download the pretrained wav2vec2 model. It can be local or online.
download_file(pretrained_path, save_path)
# During pretraining dropout might be set to 0. However, we might want
# to apply dropout when fine-tuning on a downstream task. Hence we need
# to modify the fairseq cfg to activate dropout (if requested).
overrides = {}
if not freeze and dropout is not None:
overrides["model"] = {}
if dropout is not None:
overrides["model"]["dropout"] = dropout
overrides["model"]["dropout_input"] = dropout
overrides["model"]["attention_dropout"] = dropout
if layer_drop is not None:
overrides["model"]["layer_drop"] = layer_drop
(
model,
cfg,
task,
) = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[save_path], arg_overrides=overrides
)
# wav2vec pretrained models may need the input waveform to be normalized
# Hence, we check if the model has be trained with or without it.
# If the information isn't contained in the checkpoint IT HAS TO BE GIVEN
# BY THE USER.
if input_norm is None:
if hasattr(cfg["task"], "normalize"):
self.normalize = cfg["task"].normalize
elif hasattr(cfg, "normalize"):
self.normalize = cfg.normalize
else:
self.normalize = False
else:
self.normalize = input_norm
model = model[0]
self.model = model
self.freeze = freeze
self.output_norm = output_norm
self.freeze_feature_extractor = freeze_feature_extractor
if self.freeze:
logger.warning(
"speechbrain.lobes.models.fairseq_wav2vec - wav2vec 2.0 is frozen."
)
self.model.eval()
# Freeze parameters
for param in self.model.parameters():
param.requires_grad = False
else:
self.model.train()
if self.freeze_feature_extractor:
logger.warning(
"speechbrain.lobes.models.fairseq_wav2vec - wav2vec 2.0 feature extractor is frozen."
)
self.model.feature_extractor.eval()
for param in self.model.feature_extractor.parameters():
param.requires_grad = False
# Randomly initialized layers if pretrain is False
if not (pretrain):
self.reset_layer(self.model)
# Following the fairseq implementation of downstream training,
# we remove some modules that are unnecessary.
self.remove_pretraining_modules()
[docs]
def forward(self, wav, wav_lens):
"""Takes an input waveform and return its corresponding wav2vec encoding.
Arguments
---------
wav : torch.Tensor (signal)
A batch of audio signals to transform to features.
"""
padding_mask = self.make_masks(wav, wav_len=wav_lens)
# If we freeze, we simply remove all grads and features from the graph.
if self.freeze:
with torch.no_grad():
return self.extract_features(wav, padding_mask)
return self.extract_features(wav, padding_mask)
[docs]
def reset_layer(self, model):
"""Reinitializes the parameters of the network"""
if hasattr(model, "reset_parameters"):
model.reset_parameters()
for child_layer in model.children():
if model != child_layer:
self.reset_layer(child_layer)
[docs]
def remove_pretraining_modules(self):
""" Remove uneeded modules. Inspired by the same fairseq function."""
self.model.quantizer = None
self.model.project_q = None
self.model.target_glu = None
self.model.final_proj = None
[docs]
def make_masks(self, src, wav_len=None, pad_idx=0):
"""This method generates the padding masks.
Arguments
---------
src : tensor
The sequence to the encoder (required).
wav_len : tensor
The relative length of the wav given in SpeechBrain format.
pad_idx : int
The index for <pad> token (default=0).
"""
src_key_padding_mask = None
if wav_len is not None:
abs_len = torch.round(wav_len * src.shape[1])
src_key_padding_mask = ~length_to_mask(abs_len).bool()
return src_key_padding_mask
[docs]
class FairseqWav2Vec1(nn.Module):
"""This lobes enables the integration of fairseq pretrained wav2vec1.0 models.
Arguments
---------
pretrained_path : str
Path of the pretrained wav2vec1 model. It can be a url or a local path.
save_path : str
Path and filename of the downloaded model.
output_norm : bool (default: True)
If True, a layer_norm (affine) will be applied to the output obtained
from the wav2vec model.
freeze : bool (default: True)
If True, the model is frozen. If False, the model will be trained
alongside with the rest of the pipeline.
pretrain : bool (default: True)
If True, the model is pretrained with the specified source.
If False, the randomly-initialized model is instantiated.
Example
-------
>>> inputs = torch.rand([10, 600])
>>> model_url = ""
>>> save_path = "models_checkpoints/wav2vec.pt"
>>> model = FairseqWav2Vec1(model_url, save_path)
>>> outputs = model(inputs)
>>> outputs.shape
torch.Size([10, 100, 512])
"""
def __init__(
self,
pretrained_path,
save_path,
output_norm=True,
freeze=True,
pretrain=True,
):
super().__init__()
self.freeze = freeze
self.output_norm = output_norm
# Download the pretrained wav2vec1 model. It can be local or online.
download_file(pretrained_path, save_path)
(
model,
cfg,
task,
) = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[pretrained_path]
)
self.model = model
self.model = self.model[0]
if self.freeze:
self.model.eval()
# Randomly initialized layers if pretrain is False
if not (pretrain):
self.reset_layer(self.model)
[docs]
def forward(self, wav):
"""Takes an input waveform and return its corresponding wav2vec encoding.
Arguments
---------
wav : torch.Tensor (signal)
A batch of audio signals to transform to features.
"""
# If we freeze, we simply remove all grads and features from the graph.
if self.freeze:
with torch.no_grad():
return self.extract_features(wav).detach()
return self.extract_features(wav)
[docs]
def reset_layer(self, model):
"""Reinitializes the parameters of the network"""
if hasattr(model, "reset_parameters"):
model.reset_parameters()
for child_layer in model.children():
if model != child_layer:
self.reset_layer(child_layer)