app.py

import streamlit as st
import torch
import librosa
from transformers import HubertForSequenceClassification, Wav2Vec2FeatureExtractor
from transformers import pipeline

# Title of the app
st.title("Emotion Recognition from Speech")

# Upload audio file
uploaded_file = st.file_uploader("Choose an audio file...", type=["wav"])

# Load the model and feature extractor
model = HubertForSequenceClassification.from_pretrained("superb/hubert-large-superb-er")
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("superb/hubert-large-superb-er")
classifier = pipeline("audio-classification", model="superb/hubert-large-superb-er")

if uploaded_file is not None:
    # Load and preprocess audio file
    speech, sr = librosa.load(uploaded_file, sr=16000, mono=True)
    
    # Display audio player
    st.audio(uploaded_file, format='audio/wav')
    
    # Process the audio
    inputs = feature_extractor(speech, sampling_rate=16000, padding=True, return_tensors="pt")
    
    # Predict emotion
    with torch.no_grad():
        logits = model(**inputs).logits
        predicted_ids = torch.argmax(logits, dim=-1)
        labels = [model.config.id2label[_id] for _id in predicted_ids.tolist()]
    
    # Display the result
    # st.write("Predicted Emotion:", labels[0])
    
    # Alternatively using the pipeline
    results = classifier(uploaded_file, top_k=5)
    st.write("Top 5 Predicted Emotions:")
    for result in results:
        st.write(f"{result['label']}: {result['score']:.4f}")