app.py

import os
from fastai.vision.all import *
import gradio as gr
import pickle
import tempfile
from transformers import AutoTokenizer, AutoModelWithLMHead
from speechbrain.inference.interfaces import foreign_class



# Facial expression classifier

# Emotion
learn_emotion = load_learner('emotions_vgg.pkl')
learn_emotion_labels = learn_emotion.dls.vocab


# Predict
def predict(img):
    img = PILImage.create(img)
    pred_emotion, pred_emotion_idx, probs_emotion = learn_emotion.predict(img)
    predicted_emotion = learn_emotion_labels[pred_emotion_idx]
    return predicted_emotion


# Gradio
title = "Facial Emotion Detector"

description = gr.Markdown(
                """Ever wondered what a person might be feeling looking at their picture?
                 Well, now you can! Try this fun app. Just upload a facial image in JPG or
                 PNG format. You can now see what they might have felt when the picture
                 was taken.

                 **Tip**: Be sure to only include face to get best results. Check some sample images
                 below for inspiration!""").value

article = gr.Markdown(
             """**DISCLAIMER:** This model does not reveal the actual emotional state of a person. Use and
             interpret results at your own risk!. 

             **PREMISE:** The idea is to determine an overall emotion of a person
             based on the pictures. We are restricting pictures to only include close-up facial
             images.

             **DATA:** FER2013 dataset consists of 48x48 pixel grayscale images of faces.Images
             are assigned one of the 7 emotions: Angry, Disgust, Fear, Happy, Sad, Surprise, and Neutral.
             
              """).value

enable_queue=True

examples = ['happy1.jpg', 'happy2.jpg', 'angry1.png', 'angry2.jpg', 'neutral1.jpg', 'neutral2.jpg']

image_mode=gr.Interface(fn = predict, 
             inputs = gr.Image( image_mode='L'), 
             outputs = [gr.Label(label='Emotion')], #gr.Label(),
             title = title,
             examples = examples,
             description = description,
             article=article,
             allow_flagging='never')




# Txet Model

# Load tokenizer and model from pickles
with open("emotion_tokenizer.pkl", "rb") as f:
    tokenizer = pickle.load(f)

with open("emotion_model.pkl", "rb") as f:
    model = pickle.load(f)



def classify_emotion(text):
    # Tokenize input text and generate output
    input_ids = tokenizer.encode("emotion: " + text, return_tensors="pt")
    output = model.generate(input_ids)
    output_text = tokenizer.decode(output[0], skip_special_tokens=True)
    
    # Classify the emotion into positive, negative, or neutral
    if output_text in ["joy", "love"]:
        return "Positive"
    elif output_text == "surprise":
        return "Neutral"
    else:
        return "Negative"
    return output_text


text_model = gr.Interface(fn=classify_emotion, inputs="textbox", outputs="textbox")









# Initialize the classifier
classifier = foreign_class(source="speechbrain/emotion-recognition-wav2vec2-IEMOCAP", pymodule_file="custom_interface.py", classname="CustomEncoderWav2vec2Classifier")

def save_uploaded_file(uploaded_file):
    temp_dir = tempfile.TemporaryDirectory()
    file_path = os.path.join(temp_dir.name, uploaded_file.name)
    with open(file_path, "wb") as f:
        f.write(uploaded_file.getbuffer())
    return file_path


def emotion(file_path):
   
    if file_path:
        # Classify the file
        out_prob, score, index, text_lab = classifier.classify_file(file_path)
        if isinstance(text_lab, list):
            text_lab = text_lab[0]
        # Map the original labels to the desired categories
        emotion_mapping = {
            'neu': 'Neutral',
            'ang': 'Angry',
            'hap': 'Happy',
            'sad': 'Sadness'
        }
        # Get the corresponding category from the mapping
        emotion_category = emotion_mapping.get(text_lab, 'Unknown')
        
        emotion_category = emotion_mapping.get(text_lab, 'Unknown')
        # Return the emotion category
        return emotion_category
    else:
        return "Please provide the path to an audio file."





audio_model = gr.Interface(fn=emotion, inputs="textbox", outputs="textbox")







main_model = gr.TabbedInterface([text_model, image_mode,audio_model], ["Text Emotion Recognition", "Image Emotion Recognition" , "Audio Emotion Recognition"])

main_model.launch()