update LeBenchmark

README.md

@@ -37,7 +37,7 @@ The performance of the model is the following:
 This ASR system is composed of 2 different but linked blocks:
 - Tokenizer (unigram) that transforms words into subword units and trained with
 the train transcriptions (train.tsv) of CommonVoice (FR).
-- Acoustic model (wav2vec2.0 + CTC/Attention). A pretrained wav2vec 2.0 model ([wav2vec2-large-xlsr-53-french](https://huggingface.co/facebook/wav2vec2-large-xlsr-53-french)) is combined with two DNN layers and finetuned on CommonVoice FR. 
+- Acoustic model (wav2vec2.0 + CTC/Attention). A pretrained wav2vec 2.0 model ([LeBenchmark/wav2vec2-FR-M-large](https://huggingface.co/LeBenchmark/wav2vec2-FR-M-large)) is combined with two DNN layers and finetuned on CommonVoice FR. 
 The obtained final acoustic representation is given to the CTC and attention decoders.
 
 

asr.ckpt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee40bc648d23dccd4d6d8cf77eb317aede679218ad192c96ad631921e7561024
+oid sha256:8f9b32cfe8a7d10fa852874b5507661d95cdb8c9c8dd9add45976e786e08c52e
 size 60570064

config.json

@@ -1,8 +1,8 @@
 {
-  "activation_dropout": 0.1,
+  "activation_dropout": 0.0,
   "apply_spec_augment": true,
   "architectures": [
-    "Wav2Vec2ForCTC"
+    "Wav2Vec2Model"
   ],
   "attention_dropout": 0.1,
   "bos_token_id": 1,
@@ -42,20 +42,27 @@
   "feat_extract_dropout": 0.0,
   "feat_extract_norm": "layer",
   "feat_proj_dropout": 0.1,
-  "final_dropout": 0.1,
+  "final_dropout": 0.0,
   "gradient_checkpointing": false,
   "hidden_act": "gelu",
   "hidden_dropout": 0.1,
-  "hidden_dropout_prob": 0.1,
   "hidden_size": 1024,
   "initializer_range": 0.02,
   "intermediate_size": 4096,
   "layer_norm_eps": 1e-05,
   "layerdrop": 0.1,
+  "mask_channel_length": 10,
+  "mask_channel_min_space": 1,
+  "mask_channel_other": 0.0,
+  "mask_channel_prob": 0.0,
+  "mask_channel_selection": "static",
   "mask_feature_length": 10,
   "mask_feature_prob": 0.0,
   "mask_time_length": 10,
-  "mask_time_prob": 0.05,
+  "mask_time_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.075,
+  "mask_time_selection": "static",
   "model_type": "wav2vec2",
   "num_attention_heads": 16,
   "num_conv_pos_embedding_groups": 16,
@@ -63,6 +70,6 @@
   "num_feat_extract_layers": 7,
   "num_hidden_layers": 24,
   "pad_token_id": 0,
-  "transformers_version": "4.4.0.dev0",
-  "vocab_size": 49
+  "transformers_version": "4.5.1",
+  "vocab_size": 32
 }

hyperparams.yaml

@@ -5,7 +5,7 @@
 # ################################
 
 sample_rate: 16000
-wav2vec2_hub: facebook/wav2vec2-large-xlsr-53-french
+wav2vec2_hub: LeBenchmark/wav2vec2-FR-M-large
 
 # BPE parameters
 token_type: unigram  # ["unigram", "bpe", "char"]

wav2vec2.ckpt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5675c122faaa76ed0e81e658a98a7bd6e498cd79f2f171b158a6dae10985c49c
+oid sha256:ae1869d41bd746312a183ce45f3119696d6a275680b0a01a7e5d2ebeba7e8a42
 size 1261930757