model upd

README.md

@@ -21,12 +21,12 @@ model-index:
     metrics:
        - name: Test WER
          type: wer
-         value: 34.38
+         value: 31.30
 ---
 
 # Wav2Vec2-Large-XLSR-53-Ukrainian
 
-Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Ukrainian using the [Common Voice](https://huggingface.co/datasets/common_voice) dataset.
+Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Ukrainian using the [Common Voice](https://huggingface.co/datasets/common_voice) and sample of [M-AILABS Ukrainian Corpus](https://www.caito.de/2019/01/the-m-ailabs-speech-dataset/) datasets.
 
 When using this model, make sure that your speech input is sampled at 16kHz.
 
@@ -118,10 +118,10 @@ print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"],
 
 ```
 
-**Test Result**: 34.38
+**Test Result**: 31.30
 
 ## Training
 
-The Common Voice `train`, `validation`.
+The Common Voice `train`, `validation` and sample of the M-AILABS Ukrainian corpus.
 
 The script used for training will be available [here](https://github.com/arampacha/hf-sprint-xlsr) soon.

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "facebook/wav2vec2-large-xlsr-53",
+  "_name_or_path": "arampacha/wav2vec2-large-xlsr-ukrainian",
   "activation_dropout": 0.1,
   "apply_spec_augment": true,
   "architectures": [
@@ -42,7 +42,7 @@
   "feat_extract_activation": "gelu",
   "feat_extract_dropout": 0.0,
   "feat_extract_norm": "layer",
-  "feat_proj_dropout": 0.008,
+  "feat_proj_dropout": 0.004,
   "final_dropout": 0.0,
   "gradient_checkpointing": true,
   "hidden_act": "gelu",

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7fb8c6fd0cf3288fa3cbe26a6906bb29a5a5eea3b237bff64b406614bfb7ed04
+oid sha256:797a5b497ee33bf44024d2596e6bebf6591ba15e6a3d18df7e147bb1ed92d1be
 size 1262118359