SabadModi/MultilingualSentimentAnalysis

1import gradio as gr2import tensorflow as tf 3import numpy as np4import librosa5import time6from transformers import pipeline7from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer8 9p = pipeline("automatic-speech-recognition")10 11def onlyEnglish(filename):12    model = tf.keras.models.load_model("models/English.hdf5")13    class_names = ['Anger','Anxious','Apologetic','Concerned','Encouraging','Excited','Happiness','Sadness']14    audio, sample_rate = librosa.load(filename, res_type='kaiser_fast') 15    mfccs_features = librosa.feature.mfcc(y=audio, sr=sample_rate, n_mfcc=40)16    mfccs_scaled_features = np.mean(mfccs_features.T,axis=0)17    mfccs_scaled_features=mfccs_scaled_features.reshape(1,-1)18    predicted_label=model.predict(mfccs_scaled_features)19    classes_x=np.argmax(predicted_label,axis=1)20    class_num = classes_x[0]21    return class_names[class_num]22 23def englishGerman(filename):24    model = tf.keras.models.load_model("models/English-German.hdf5")25    class_names = ['Anger','Happiness','Sadness']26    audio, sample_rate = librosa.load(filename, res_type='kaiser_fast') 27    mfccs_features = librosa.feature.mfcc(y=audio, sr=sample_rate, n_mfcc=40)28    mfccs_scaled_features = np.mean(mfccs_features.T,axis=0)29    mfccs_scaled_features=mfccs_scaled_features.reshape(1,-1)30    predicted_label=model.predict(mfccs_scaled_features)31    classes_x=np.argmax(predicted_label,axis=1)32    class_num = classes_x[0]33    return class_names[class_num]34 35def multiple(filename):36    model = tf.keras.models.load_model("models/Multiple.hdf5")37    class_names = ["Anger","Disgust","Fear","Happiness","Neutral","Sadness"]38    audio, sample_rate = librosa.load(filename, res_type='kaiser_fast') 39    mfccs_features = librosa.feature.mfcc(y=audio, sr=sample_rate, n_mfcc=40)40    mfccs_scaled_features = np.mean(mfccs_features.T,axis=0)41    mfccs_scaled_features=mfccs_scaled_features.reshape(1,-1)42    predicted_label=model.predict(mfccs_scaled_features)43    classes_x=np.argmax(predicted_label,axis=1)44    class_num = classes_x[0]45    return class_names[class_num]46 47#calculate the negative, positive, neutral and compound scores, plus verbal evaluation48def sentiment_vader(sentence):49 50    # Create a SentimentIntensityAnalyzer object.51    sid_obj = SentimentIntensityAnalyzer()52 53    sentiment_dict = sid_obj.polarity_scores(sentence)54    negative = sentiment_dict['neg']55    neutral = sentiment_dict['neu']56    positive = sentiment_dict['pos']57    compound = sentiment_dict['compound']58 59    if sentiment_dict['compound'] >= 0.05 :60        overall_sentiment = "Positive"61 62    elif sentiment_dict['compound'] <= - 0.05 :63        overall_sentiment = "Negative"64 65    else :66        overall_sentiment = "Neutral"67 68    return overall_sentiment69 70def transcribe(audio, state=""):71    time.sleep(3)72    text = p(audio)["text"]73    text = sentiment_vader(text)74    return text75 76with gr.Blocks() as demo:77    gr.Markdown("<h1 style='text-align: center'>Deployed 3 different Models for different types of languages.</h1><ol><li>Model 1 has 8 Classes of Emotions and is trained only on English Language. This model extracts features and trains a NN model upon that.</li><li>Model 2 has 3 Classes of Emotions and is trained only on English and German. This model extracts features and trains a NN model upon that.</li><li>Model 3 has 6 Classes of Emotions and is trained only on English, French, German and Mexican. This model extracts features and trains a NN model upon that.</li><li>Model 4 has 3 Classes of Emotions and is trained only on English. It converts Speech-to-Text and then performs Sentiment Analysis.</li></ol>")78    with gr.Tabs():79        with gr.TabItem("Only English Model"):80            english_input = gr.inputs.Audio(label="Input Audio", type="filepath")81            english_button = gr.Button("Submit")82            english_output = gr.outputs.Label(num_top_classes = 8)83            84        with gr.TabItem("English-German Model"):85            two_input = gr.inputs.Audio(label="Input Audio", type="filepath")86            two_button = gr.Button("Submit")87            two_output = gr.outputs.Label(num_top_classes = 3)88            89        with gr.TabItem("English-French-German-Mexican Model"):90            multiple_input = gr.inputs.Audio(label="Input Audio", type="filepath")91            multiple_button = gr.Button("Submit")92            multiple_output = gr.outputs.Label(num_top_classes = 6)93        94        with gr.TabItem("Speech to Text"):95            stt_input = gr.inputs.Audio(label="Input Audio", type="filepath")96            stt_button = gr.Button("Submit")97            stt_output = gr.outputs.Label(num_top_classes = 6)98 99    english_button.click(onlyEnglish, inputs=english_input, outputs=english_output)100    two_button.click(englishGerman, inputs=two_input, outputs=two_output)101    multiple_button.click(englishGerman, inputs=multiple_input, outputs=multiple_output)102    stt_button.click(transcribe, inputs=stt_input, outputs=stt_output)103 104demo.launch()105 106