Update app.py

app.py

@@ -1,34 +1,45 @@
 import gradio as gr
 import time
-from transformers import pipeline, AutoModelForSequenceClassification, AutoTokenizer
+import openai
+import json
+import os
 
-# ASR pipeline
-asr_pipeline = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h")
-
-# Load classifier model and tokenizer
-classifier_model = AutoModelForSequenceClassification.from_pretrained("Ngadou/bert-sms-spam-dectector")
-classifier_tokenizer = AutoTokenizer.from_pretrained("Ngadou/bert-sms-spam-dectector")
+openai.api_key = os.environ.get('OPENAI_KEY')
 
 def classify_audio(audio):
     # Transcribe the audio to text
-    text = asr_pipeline(audio)["text"]
+    audio_transcript = asr_pipeline(audio)["text"]
+    audio_transcript = audio_transcript.lower()
+
+    messages = [
+        {"role": "system", "content": "Is this chat a scam, spam or is safe? Only answer in JSON format with 'classification': '' as string and 'reasons': '' as the most plausible reasons why. The reason should be explaning to the potential victim why the conversation is probably a scam"},
+        {"role": "user", "content": audio_transcript},
+    ]
+
+    # Call the OpenAI API to generate a response
+    response = openai.ChatCompletion.create(
+        model="gpt-4",  # Replace with the actual GPT-4 model ID
+        messages=messages
+    )
 
-    # Tokenize the text and feed it to the model
-    inputs = classifier_tokenizer.encode_plus(text, return_tensors="pt")
-    outputs = classifier_model(**inputs)
+    # Extract the generated text
+    text = response.choices[0].message['content']
+    text = json.loads(text)
 
-    # Get the prediction (0 = ham, 1 = spam)
-    prediction = outputs.logits.argmax(dim=1).item()
+    # Get the decision and reasons from the JSON dictionary
+    decision = text["classification"]
+    reasons = text["reasons"]
 
     # Return the transcription and the prediction as a dictionary
-    return text, "Scam" if prediction == 1 else "Safe Message"
+    return audio_transcript, decision, reasons
 
 gr.Interface(
-    fn=classify_audio, 
+    fn=classify_audio,
     inputs=gr.inputs.Audio(source="upload", type="filepath"),
     outputs=[
         gr.outputs.Textbox(label="Transcription"),
         gr.outputs.Textbox(label="Classification"),
+        gr.outputs.Textbox(label="Reason"),
     ],
     live=True
-).launch(share=True)
+).launch(share=True, debug=True)