Adding initial files

app.py

@@ -0,0 +1,82 @@
+import librosa, joblib, numpy as np, gradio as gr
+from scipy.interpolate import interp1d
+from pyAudioAnalysis import ShortTermFeatures
+from pydub.silence import detect_nonsilent
+from pydub import AudioSegment
+
+
+def smart_resize(arr, target_size):
+    current_size = arr.shape[1]
+
+    current_idx = np.linspace(0, current_size - 1, current_size)
+    target_idx = np.linspace(0, current_size - 1, target_size)
+
+    # Interpolate/extrapolate
+    interp_func = interp1d(current_idx, arr.squeeze(), kind='linear', fill_value="extrapolate")
+    resized_arr = interp_func(target_idx)
+    
+    return resized_arr.reshape(1, target_size)
+
+def remove_silence(wav_file):
+    audSeg = AudioSegment.from_wav(wav_file)
+    non_silence_ranges = detect_nonsilent(audSeg, min_silence_len=5, silence_thresh=-30)
+
+    if not non_silence_ranges: 
+        sound = audSeg
+    else:
+        start = non_silence_ranges[0][0]
+        end = non_silence_ranges[-1][1]
+        trimmed_sound = audSeg[start:end]
+        sound = trimmed_sound
+
+    sound.export('audio.wav', format="wav")
+
+def transform_data(audio):
+    remove_silence(audio)
+    x, sr = librosa.load('audio.wav')
+
+    result, f_names = ShortTermFeatures.feature_extraction(x, sr, 0.050*sr, 0.025*sr)
+
+    resize_features = smart_resize(result.reshape(1,-1), 20) 
+
+    return resize_features
+
+def predict(newdf, loaded_model):
+    
+    prediction = loaded_model.predict(newdf)
+
+    return prediction 
+    
+def get_label(newpred):
+    if newpred == 0:
+        return 'No'
+    else:
+        return 'Si'
+
+def load_model():
+    ram_for = joblib.load('models/sgd_90.pkl') 
+
+    return ram_for
+    
+def main(audio):
+    newdf = transform_data(audio)
+    loaded_model = load_model()
+    newpred = predict(newdf, loaded_model)
+    final = get_label(newpred)
+    
+    return final
+
+demo = gr.Interface(
+    title = "Autoagent | YES or NO Classification - Layer7",
+    description = "<h3>This model is useful to classify if the user says 'Si' or 'No'. 🎙️ </h3> <br> <b>Record your voice:</b>",
+    allow_flagging = "never",
+    fn = main,
+    inputs = gr.Audio(
+        sources=["microphone"],
+        type="filepath",
+    ), 
+    outputs = gr.Textbox(label="Clasification")
+    )
+    
+if __name__ == "__main__":
+    demo.launch(show_api=False)  

models/sgd_90.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03f33949524bce752dae123a0fcbaac91be1e390bcaa338f141835463c795a78
+size 1248

requirements.txt

@@ -0,0 +1,7 @@
+gradio
+joblib
+numpy
+librosa
+scipy
+pyAudioAnalysis
+pydub