Upload fine_ast.py

fine_ast.py

@@ -0,0 +1,83 @@
+
+from datasets import load_dataset
+dataset = load_dataset("audiofolder", data_dir="data")
+#dataset= dataset["train"].train_test_split(seed=42, shuffle=True, test_size=0.1)
+from transformers import ASTForAudioClassification
+from transformers import ASTFeatureExtractor
+from transformers import TrainingArguments
+import numpy as np
+from transformers import Trainer
+import evaluate
+batch_size = 8
+gradient_accumulation_steps = 1
+num_train_epochs = 10
+labels=["noise","speech"]
+num_labels = 2
+max_duration = 5
+model_id="bookbot/distil-ast-audioset"
+model_name = "speechVSnoise"
+
+label2id, id2label = dict(), dict()
+for i, label in enumerate(labels):
+    label2id[label] = str(i)
+    id2label[str(i)] = label
+
+model = ASTForAudioClassification.from_pretrained(
+    model_id,
+    num_labels=num_labels,    label2id=label2id,
+    id2label=id2label,
+ignore_mismatched_sizes=True
+)
+feature_extractor = ASTFeatureExtractor.from_pretrained(
+    model_id, do_normalize=True, return_attention_mask=False
+)
+
+
+def preprocess_function(examples):
+    audio_arrays = [x["array"] for x in examples["audio"]]
+    inputs = feature_extractor(
+        audio_arrays,
+        sampling_rate=feature_extractor.sampling_rate,
+        max_length=int(feature_extractor.sampling_rate * max_duration),
+        truncation=True,
+
+    )
+    return inputs
+dataset_encoded = dataset.map(
+    preprocess_function,
+    batched=True,
+    batch_size=1674,
+    num_proc=1,
+)
+metric = evaluate.load("accuracy")
+def compute_metrics(eval_pred):
+    predictions = np.argmax(eval_pred.predictions, axis=1)
+    return metric.compute(predictions=predictions, references=eval_pred.label_ids)
+training_args = TrainingArguments(
+    f"{model_name}",
+    evaluation_strategy="epoch",
+    save_strategy="epoch",
+    learning_rate=5e-5,
+    per_device_train_batch_size=batch_size,
+    gradient_accumulation_steps=gradient_accumulation_steps,
+    per_device_eval_batch_size=batch_size,
+    num_train_epochs=num_train_epochs,
+    warmup_ratio=0.1,
+    logging_steps=5,
+    load_best_model_at_end=True,
+#    metric_for_best_model="accuracy",
+   # push_to_hub=True,
+)
+
+from transformers import Trainer
+
+trainer = Trainer(
+    model,
+    training_args,
+    train_dataset=dataset_encoded["train"],
+    eval_dataset=dataset_encoded["train"],
+    tokenizer=feature_extractor,
+ #   compute_metrics=compute_metrics,
+)
+trainer.train()
+