First version

README.md

@@ -0,0 +1,99 @@
+---
+language: sl
+datasets:
+- common_voice
+tags:
+- audio
+- automatic-speech-recognition
+- speech
+- xlsr-fine-tuning-week
+license: apache-2.0
+model-index:
+- name: XLSR Wav2Vec2 Slovene
+  results:
+  - task: 
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    dataset:
+      name: Common Voice sl
+      type: common_voice
+      args: sl
+    metrics:
+       - name: Test WER
+         type: wer
+         value: 36.97
+---
+# Wav2Vec2-Large-XLSR-53-Slovene
+Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Slovene using the [Common Voice](https://huggingface.co/datasets/common_voice)
+When using this model, make sure that your speech input is sampled at 16kHz.
+
+## Usage
+The model can be used directly (without a language model) as follows:
+
+```python
+import torch
+import torchaudio
+from datasets import load_dataset
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+test_dataset = load_dataset("common_voice", "sl", split="test[:2%]").
+processor = Wav2Vec2Processor.from_pretrained("mrshu/wav2vec2-large-xlsr-slovene")
+model = Wav2Vec2ForCTC.from_pretrained("mrshu/wav2vec2-large-xlsr-slovene")
+resampler = torchaudio.transforms.Resample(48_000, 16_000)
+# Preprocessing the datasets.
+# We need to read the aduio files as arrays
+def speech_file_to_array_fn(batch):
+	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)
+inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
+with torch.no_grad():
+	logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
+predicted_ids = torch.argmax(logits, dim=-1)
+print("Prediction:", processor.batch_decode(predicted_ids))
+print("Reference:", test_dataset["sentence"][:2])
+```
+
+
+## Evaluation
+The model can be evaluated as follows on the Slovene test data of Common Voice.
+```python
+import torch
+import torchaudio
+from datasets import load_dataset, load_metric
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+import re
+test_dataset = load_dataset("common_voice", "sl", split="test")
+wer = load_metric("wer")
+processor = Wav2Vec2Processor.from_pretrained("mrshu/wav2vec2-large-xlsr-slovene")
+model = Wav2Vec2ForCTC.from_pretrained("mrshu/wav2vec2-large-xlsr-slovene")
+model.to("cuda")
+chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\%\‘\”\�\«\»\)\(\„\'\–\’\—]'
+resampler = torchaudio.transforms.Resample(48_000, 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()
+	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
+result = test_dataset.map(evaluate, batched=True, batch_size=8)
+print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
+```
+
+**Test Result**: 36.97 %
+
+## Training
+The Common Voice `train`, `validation` datasets were used for training.
+The script used for training can be found [here](https://colab.research.google.com/drive/14uahdilysnFsiYniHxY9fyKjFGuYQe7p)
+

config.json

@@ -0,0 +1,76 @@
+{
+  "_name_or_path": "/content/drive/MyDrive/wav2vec2-large-xlsr-slovene-demo/checkpoint-2400",
+  "activation_dropout": 0.0,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": false,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.0,
+  "final_dropout": 0.0,
+  "gradient_checkpointing": true,
+  "hidden_act": "gelu",
+  "hidden_dropout": 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_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.05,
+  "mask_time_selection": "static",
+  "model_type": "wav2vec2",
+  "num_attention_heads": 16,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "pad_token_id": 30,
+  "transformers_version": "4.4.0",
+  "vocab_size": 31
+}

optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6412a776d986b7778ed3bc3e5aeebce7da3725634501e8152750a644f8856c5
+size 2524030271

preprocessor_config.json

@@ -0,0 +1,8 @@
+{
+  "do_normalize": true,
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "return_attention_mask": true,
+  "sampling_rate": 16000
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f5876d15d270501f95c5b012a92adcc065d2e36cdcd9203dcdc815d3e22a4b6
+size 1262060951

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d925a061b55ae2fd0dc674f239e8a7c89ec7351e753745cdfa70b1eb43df83e5
+size 623

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "[UNK]", "pad_token": "[PAD]"}

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "[UNK]", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|"}

trainer_state.json

@@ -0,0 +1,100 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 29.625766871165645,
+  "global_step": 2400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 4.93,
+      "learning_rate": 6.400000000000001e-05,
+      "loss": 0.0363,
+      "step": 400
+    },
+    {
+      "epoch": 4.93,
+      "eval_loss": 0.6197775602340698,
+      "eval_runtime": 84.7989,
+      "eval_samples_per_second": 10.389,
+      "eval_wer": 0.5268051434223541,
+      "step": 400
+    },
+    {
+      "epoch": 9.87,
+      "learning_rate": 6.756476683937824e-05,
+      "loss": 0.0221,
+      "step": 800
+    },
+    {
+      "epoch": 9.87,
+      "eval_loss": 0.7040311694145203,
+      "eval_runtime": 86.0378,
+      "eval_samples_per_second": 10.24,
+      "eval_wer": 0.5272007912957468,
+      "step": 800
+    },
+    {
+      "epoch": 14.81,
+      "learning_rate": 5.098445595854923e-05,
+      "loss": 0.0174,
+      "step": 1200
+    },
+    {
+      "epoch": 14.81,
+      "eval_loss": 0.7597708702087402,
+      "eval_runtime": 85.9482,
+      "eval_samples_per_second": 10.25,
+      "eval_wer": 0.5339268051434224,
+      "step": 1200
+    },
+    {
+      "epoch": 19.75,
+      "learning_rate": 3.440414507772021e-05,
+      "loss": 0.0126,
+      "step": 1600
+    },
+    {
+      "epoch": 19.75,
+      "eval_loss": 0.7824531197547913,
+      "eval_runtime": 85.9896,
+      "eval_samples_per_second": 10.245,
+      "eval_wer": 0.5224530168150346,
+      "step": 1600
+    },
+    {
+      "epoch": 24.69,
+      "learning_rate": 1.7823834196891192e-05,
+      "loss": 0.0125,
+      "step": 2000
+    },
+    {
+      "epoch": 24.69,
+      "eval_loss": 0.769702672958374,
+      "eval_runtime": 86.2655,
+      "eval_samples_per_second": 10.213,
+      "eval_wer": 0.5224530168150346,
+      "step": 2000
+    },
+    {
+      "epoch": 29.63,
+      "learning_rate": 1.2435233160621762e-06,
+      "loss": 0.0167,
+      "step": 2400
+    },
+    {
+      "epoch": 29.63,
+      "eval_loss": 0.7395063638687134,
+      "eval_runtime": 83.5861,
+      "eval_samples_per_second": 10.54,
+      "eval_wer": 0.5198813056379822,
+      "step": 2400
+    }
+  ],
+  "max_steps": 2430,
+  "num_train_epochs": 30,
+  "total_flos": 9.370015200295258e+18,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0937970b40987b25b7283a4c90e818b7ccd2993130e0236b66f7b493b31db24f
+size 2287

vocab.json

@@ -0,0 +1 @@
+{"r": 0, "o": 1, "x": 2, "e": 3, "u": 4, "š": 5, "v": 6, "ž": 7, "c": 8, "h": 9, "i": 10, "p": 11, "č": 12, "w": 13, "b": 14, "y": 15, "s": 16, "d": 17, "m": 18, "z": 19, "l": 20, "f": 21, "n": 23, "j": 24, "a": 25, "k": 26, "t": 27, "g": 28, "|": 22, "[UNK]": 29, "[PAD]": 30}