Time stamps for CTC models

docs/source/model_doc/wav2vec2.mdx

@@ -45,6 +45,8 @@ This model was contributed by [patrickvonplaten](https://huggingface.co/patrickv
 [[autodoc]] Wav2Vec2CTCTokenizer
     - __call__
     - save_vocabulary
+    - decode
+    - batch_decode
 
 ## Wav2Vec2FeatureExtractor
 

src/transformers/models/hubert/configuration_hubert.py

@@ -14,6 +14,8 @@
 # limitations under the License.
 """ Hubert model configuration"""
 
+import math
+
 from ...configuration_utils import PretrainedConfig
 from ...utils import logging
 
@@ -248,3 +250,7 @@ def __init__(
         # ctc loss
         self.ctc_loss_reduction = ctc_loss_reduction
         self.ctc_zero_infinity = ctc_zero_infinity
+
+    @property
+    def inputs_to_logits_ratio(self):
+        return math.prod(self.conv_stride)

src/transformers/models/sew/configuration_sew.py

@@ -14,6 +14,8 @@
 # limitations under the License.
 """ SEW model configuration"""
 
+import math
+
 from ...configuration_utils import PretrainedConfig
 from ...utils import logging
 
@@ -243,3 +245,7 @@ def __init__(
         # sequence classification
         self.use_weighted_layer_sum = use_weighted_layer_sum
         self.classifier_proj_size = classifier_proj_size
+
+    @property
+    def inputs_to_logits_ratio(self):
+        return math.prod(self.conv_stride)

src/transformers/models/sew_d/configuration_sew_d.py

@@ -14,6 +14,8 @@
 # limitations under the License.
 """ SEW-D model configuration"""
 
+import math
+
 from ...configuration_utils import PretrainedConfig
 from ...utils import logging
 
@@ -279,3 +281,7 @@ def __init__(
         # sequence classification
         self.use_weighted_layer_sum = use_weighted_layer_sum
         self.classifier_proj_size = classifier_proj_size
+
+    @property
+    def inputs_to_logits_ratio(self):
+        return math.prod(self.conv_stride)

src/transformers/models/unispeech/configuration_unispeech.py

@@ -14,6 +14,8 @@
 # limitations under the License.
 """ UniSpeech model configuration"""
 
+import math
+
 from ...configuration_utils import PretrainedConfig
 from ...utils import logging
 
@@ -289,3 +291,7 @@ def __init__(
 
         # pretraining loss
         self.replace_prob = replace_prob
+
+    @property
+    def inputs_to_logits_ratio(self):
+        return math.prod(self.conv_stride)

src/transformers/models/unispeech_sat/configuration_unispeech_sat.py

@@ -14,6 +14,8 @@
 # limitations under the License.
 """ UniSpeechSat model configuration"""
 
+import math
+
 from ...configuration_utils import PretrainedConfig
 from ...utils import logging
 
@@ -306,3 +308,7 @@ def __init__(
         self.tdnn_kernel = list(tdnn_kernel)
         self.tdnn_dilation = list(tdnn_dilation)
         self.xvector_output_dim = xvector_output_dim
+
+    @property
+    def inputs_to_logits_ratio(self):
+        return math.prod(self.conv_stride)

src/transformers/models/wav2vec2/configuration_wav2vec2.py

@@ -14,6 +14,8 @@
 # limitations under the License.
 """ Wav2Vec2 model configuration"""
 
+import math
+
 from ...configuration_utils import PretrainedConfig
 from ...utils import logging
 
@@ -329,3 +331,7 @@ def __init__(
         self.tdnn_kernel = list(tdnn_kernel)
         self.tdnn_dilation = list(tdnn_dilation)
         self.xvector_output_dim = xvector_output_dim
+
+    @property
+    def inputs_to_logits_ratio(self):
+        return math.prod(self.conv_stride)