updates model

README.md

@@ -25,10 +25,10 @@ model-index:
     metrics:
        - name: Test WER
          type: wer
-         value: 23.298429
+         value: 20.220256
        - name: Test CER
          type: cer
-         value: 7.867248
+         value: 6.924060
 ---
 
 # Wav2Vec2-Large-XLSR-53-Swedish
@@ -85,11 +85,11 @@ from datasets import load_dataset, load_metric
 from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
 import re
 
-test_dataset = load_dataset("common_voice", "sv-SE", split="test") #TODO: replace {lang_id} in your language code here. Make sure the code is one of the *ISO codes* of [this](https://huggingface.co/languages) site.
+test_dataset = load_dataset("common_voice", "sv-SE", split="test")
 wer = load_metric("wer")
 
-processor = Wav2Vec2Processor.from_pretrained("vasilis/wav2vec2-large-xlsr-53-swedish") #TODO: replace {model_id} with your model id. The model id consists of {your_username}/{your_modelname}, *e.g.* `elgeish/wav2vec2-large-xlsr-53-arabic`
-model = Wav2Vec2ForCTC.from_pretrained("vasilis/wav2vec2-large-xlsr-53-swedish") #TODO: replace {model_id} with your model id. The model id consists of {your_username}/{your_modelname}, *e.g.* `elgeish/wav2vec2-large-xlsr-53-arabic`
+processor = Wav2Vec2Processor.from_pretrained("vasilis/wav2vec2-large-xlsr-53-swedish")
+model = Wav2Vec2ForCTC.from_pretrained("vasilis/wav2vec2-large-xlsr-53-swedish")
 model.to("cuda")
 
 chars_to_ignore_regex = "[\,\?\.\!\-\;\:\"\“\%\‘\”\�\']"  # TODO: adapt this list to include all special characters you removed from the data
@@ -100,36 +100,24 @@ resampler = {
     32000: torchaudio.transforms.Resample(32000, 16_000)
 }
 
-
 # Preprocessing the datasets.
 # We need to read the aduio files as arrays
 def speech_file_to_array_fn(batch):
     batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
+    for key, value in replacements.items():
+        batch["sentence"] = batch["sentence"].replace(key, value)
     speech_array, sampling_rate = torchaudio.load(batch["path"])
     batch["speech"] = resampler[sampling_rate](speech_array).squeeze().numpy()
     return batch
 
-
-test_dataset = test_dataset.map(speech_file_to_array_fn)
-
-# Preprocessing the datasets.
-# We need to read the aduio files as arrays
-def speech_file_to_array_fn(batch):
-    batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
-    speech_array, sampling_rate = torchaudio.load(batch["path"])
-    batch["speech"] = resampler(speech_array).squeeze().numpy()
-    return batch
-
 test_dataset = test_dataset.map(speech_file_to_array_fn)
 
 # Preprocessing the datasets.
 # We need to read the aduio files as arrays
 def evaluate(batch):
     inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
-
     with torch.no_grad():
         logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
-
     pred_ids = torch.argmax(logits, dim=-1)
     batch["pred_strings"] = processor.batch_decode(pred_ids)
     return batch
@@ -141,16 +129,19 @@ print("CER: {:2f}".format(100 * wer.compute(predictions=[" ".join(list(entry)) f
 
 ```
 
-**Test Result**:  23.298429 %
+**Test Result**:  20.220256 %
 
 
 ## Training
 
-
-The Common Voice train dataset was used for training. From NST
-the training split (60%) as can be found [here](https://github.com/se-asr/nst/tree/master/splits/60-20-20-80096) was used. On top of this
-a big chunk of the dataset was filtered out based on this mask
+As first step used Common Voice train dataset and parts from NST
+as can be found [here](https://github.com/se-asr/nst/tree/master).
+Part of NST where removed using this mask
 
 ```python
 mask = [(5 < len(x.split()) < 20) and np.average([len(entry) for entry in x.split()]) > 5 for x in dataset['transcript'].tolist()]
-```
+```
+
+After training like this for 20000 steps the model was finetuned on the training and validation sets of Common Voice
+only for 3500 steps more.
+

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "facebook/wav2vec2-large-xlsr-53",
+  "_name_or_path": "/speech-data-1/dev/hugging_face_finetuning_week/sw_demo/checkpoints/2020_24_3_v24/checkpoint-19600",
   "activation_dropout": 0.07,
   "apply_spec_augment": true,
   "architectures": [

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2b0a008e1c06230fd9aaf96e7c8f9e5edbbca5786d543bb6653dab60e268123
+oid sha256:9b7be02b3d56d35d6bbdff2e511de5f2b3d44e66874958aca5617d0e594a3e20
 size 1262065047