Update app.py

app.py

@@ -23,13 +23,6 @@ from transformers import AutoModelForSequenceClassification, pipeline, WhisperTo
 
 model_cache = {}
 
-# Building prediction function for gradio
-emo_dict = {
-    'sad': 'Sad', 
-    'hap': 'Happy',
-    'ang': 'Anger',
-    'neu': 'Neutral'
-}
 
 # static classes for now, but it would be best ot have the user select from multiple, and to enter their own
 class_options = {
@@ -57,12 +50,14 @@ def slider_logic(slider):
     return threshold
 
 # Create a Gradio interface with audio file and text inputs
-def classify_toxicity(audio_file, slider):
+def classify_toxicity(audio_file, selected_sounds, slider):
     # Transcribe the audio file using Whisper ASR
     if audio_file != None:
         transcribed_text = pipe(audio_file)["text"]
     else:
         transcribed_text = text_input
+
+    selected_class_names = selected_sounds.split(",")
     
     threshold = slider_logic(slider)
     model = whisper.load_model("large")
@@ -95,6 +90,20 @@ def classify_toxicity(audio_file, slider):
     )
     average_logprobs -= internal_lm_average_logprobs
     scores = average_logprobs.softmax(-1).tolist()
+    
+    class_score_dict = {class_name: score for class_name, score in zip(class_names, scores)}
+    for selected_class_name in selected_class_names:
+        if selected_class_name in class_score_dict:
+            score = class_score_dict[selected_class_name]
+            if score > threshold:
+                print(f"Threshold exceeded for class '{selected_class_name}': Score = {score:.4f}")
+
+
+    '''
+    for class_name, score in class_score_dict.items():
+        if score > threshold:
+            print(f"Threshold exceeded for class '{class_name}': Score = {score:.4f}")
+    '''   
     holder1 = {class_name: score for class_name, score in zip(class_names, scores)}
     # miso_label_dict = {label: score for label, score in classify_anxiety[0].items()}
     holder2 = ""
@@ -114,6 +123,7 @@ def positive_affirmations():
 with gr.Blocks() as iface:
     show_state = gr.State([])
     with gr.Column():
+        miso_sounds = gr.CheckboxGroup(["chewing", "breathing", "mouthsounds", "popping", "sneezing", "yawning", "smacking", "sniffling", "panting"])
         sense_slider = gr.Slider(minimum=1, maximum=5, step=1.0, label="How readily do you want the tool to intervene? 1 = in extreme cases and 5 = at every opportunity")
     with gr.Column():
         aud_input = gr.Audio(source="upload", type="filepath", label="Upload Audio File")
@@ -121,6 +131,6 @@ with gr.Blocks() as iface:
     with gr.Column():
         # out_val = gr.Textbox()
         out_class = gr.Label()
-    submit_btn.click(fn=classify_toxicity, inputs=[aud_input, sense_slider], outputs=out_class)
+    submit_btn.click(fn=classify_toxicity, inputs=[aud_input, miso_sounds, sense_slider], outputs=out_class)
 
 iface.launch()