import gradio as gr
import numpy as np
import joblib
import librosa
from sklearn.preprocessing import StandardScaler

model = joblib.load('UTI.pkl')

def predictor(audio_filename):
    y, sr = librosa.load(audio_filename, mono=True, duration=5)

    chroma_stft = np.mean(librosa.feature.chroma_stft(y=y, sr=sr))
    rmse = np.mean(librosa.feature.rms(y=y))
    spec_cent = np.mean(librosa.feature.spectral_centroid(y=y, sr=sr))
    spec_bw = np.mean(librosa.feature.spectral_bandwidth(y=y, sr=sr))
    rolloff = np.mean(librosa.feature.spectral_rolloff(y=y, sr=sr))
    zcr = np.mean(librosa.feature.zero_crossing_rate(y))
    mfcc = librosa.feature.mfcc(y=y, sr=sr)
    v = []
    for e in mfcc:
        v.append(np.mean(e))
    mfcc1 = v[0]
    mfcc2 = v[1]
    mfcc3 = v[2]
    mfcc4 = v[3]
    mfcc5 = v[4]
    mfcc6 = v[5]
    mfcc7 = v[6]
    mfcc8 = v[7]
    mfcc9 = v[8]
    mfcc10 = v[9]
    mfcc11 = v[10]
    mfcc12 = v[11]
    mfcc13 = v[12]
    mfcc14 = v[13]
    mfcc15 = v[14]
    mfcc16 = v[15]
    mfcc17 = v[16]
    mfcc18 = v[17]
    mfcc19 = v[18]
    mfcc20 = v[19]

    features = np.array([[chroma_stft,rmse,spec_cent,spec_bw,rolloff,zcr,mfcc1,mfcc2,mfcc3,mfcc4,mfcc5,mfcc6,mfcc7,mfcc8,mfcc9,mfcc10,mfcc11,mfcc12,mfcc13,mfcc14,mfcc15,mfcc16,mfcc17,mfcc18,mfcc19,mfcc20]])

    prediction = model.predict(StandardScaler().fit_transform(features))
    
    if prediction[0] == 0:
        result = 'Normal'
    else:
        result = 'Infected'
    return result

outputs = gr.outputs.Textbox()
app = gr.Interface(fn=predictor, inputs='text', outputs=outputs,description="UTI Prediction Model")
app.launch()