Initial commit

app.py

@@ -0,0 +1,42 @@
+import streamlit as st
+from transformers import HubertForSequenceClassification, HubertConfig, Wav2Vec2FeatureExtractor
+import torch
+import soundfile as sf
+
+# Load model and tokenizer
+model_name = "model_hubert_finetuned_nopeft.pth"  # Replace with your model path or Hugging Face model hub path
+config = HubertConfig.from_pretrained(model_name)
+config.id2label = {0: 'neu', 1: 'hap', 2: 'ang', 3: 'sad', 4: 'dis', 5: 'sur', 6: 'fea', 7: 'cal'}
+config.label2id = {"neu": 0, "hap": 1, "ang": 2, "sad": 3, "dis": 4, "sur": 5, "fea": 6, "cal": 7}
+config.num_labels = 8  # Set it to the number of classes in your SER task
+
+# Load the pre-trained model with the modified configuration
+model = HubertForSequenceClassification.from_pretrained(model_name, config=config, ignore_mismatched_sizes=True)
+model.to('cuda' if torch.cuda.is_available() else 'cpu')
+model.eval()
+
+# Load feature extractor
+feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("superb/hubert-large-superb-er")
+
+st.title("Speech Emotion Recognition Model")
+
+uploaded_file = st.file_uploader("Upload an audio file", type=["wav"])
+
+if uploaded_file is not None:
+    # Load audio file
+    audio_input, sampling_rate = sf.read(uploaded_file)
+    
+    # Preprocess audio input
+    inputs = feature_extractor(audio_input, sampling_rate=sampling_rate, return_tensors="pt", padding=True)
+    inputs = {key: value.to('cuda' if torch.cuda.is_available() else 'cpu') for key, value in inputs.items()}
+    
+    # Get prediction
+    with torch.no_grad():
+        outputs = model(**inputs)
+        logits = outputs.logits
+        probabilities = torch.softmax(logits, dim=-1)
+        predicted_class = torch.argmax(probabilities, dim=1).item()
+    
+    # Display prediction
+    st.write(f"Predicted class: {config.id2label[predicted_class]}")
+    st.write(f"Class probabilities: {probabilities}")

requirements.txt

@@ -0,0 +1,4 @@
+torch
+transformers
+soundfile
+streamlit