Update src/streamlit_app.py

src/streamlit_app.py

@@ -10,10 +10,18 @@ os.environ["HF_HOME"] = "/app/.cache/huggingface"
 os.environ["TORCH_HOME"] = "/app/.cache/torch"
 hf_token = os.getenv("HateSpeechMujtabatoken")
 
-whisper_processor = WhisperProcessor.from_pretrained("openai/whisper-tiny", token=hf_token)
-whisper_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny", token=hf_token)
-text_model = AutoModelForSequenceClassification.from_pretrained("Hate-speech-CNERG/bert-base-uncased-hatexplain", token=hf_token)
-tokenizer = AutoTokenizer.from_pretrained("Hate-speech-CNERG/bert-base-uncased-hatexplain", token=hf_token)
+import torch
+import torchaudio
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+from transformers import WhisperProcessor, WhisperForConditionalGeneration
+import streamlit as st
+
+whisper_processor = WhisperProcessor.from_pretrained("Hate-speech-CNERG/bert-base-uncased-hatexplain")
+whisper_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small")
+text_model = AutoModelForSequenceClassification.from_pretrained("Hate-speech-CNERG/bert-base-uncased-hatexplain")
+tokenizer = AutoTokenizer.from_pretrained("Hate-speech-CNERG/bert-base-uncased-hatexplain")
+
+label_map = {0: "Not Hate Speech", 1: "Hate Speech"}
 
 def transcribe(audio_path):
     waveform, sample_rate = torchaudio.load(audio_path)
@@ -23,31 +31,34 @@ def transcribe(audio_path):
     return transcription
 
 def extract_text_features(text):
-    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
+    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)
     outputs = text_model(**inputs)
-    predicted_class = outputs.logits.argmax(dim=1).item()
-    return "Hate Speech" if predicted_class == 1 else "Not Hate Speech"
-
-def predict(audio_file, text_input):
-    if not audio_file and not text_input:
-        return "Please provide either an audio file or some text."
-    if audio_file:
-        audio_path = "temp_audio.wav"
-        with open(audio_path, "wb") as f:
-            f.write(audio_file.read())
-        transcribed_text = transcribe(audio_path)
-        prediction = extract_text_features(text_input or transcribed_text)
-        if text_input:
-            return f"Predicted: {prediction}"
-        else:
-            return f"Predicted: {prediction} \n\n(Transcribed: {transcribed_text})"
+    pred_label = outputs.logits.argmax(dim=1).item()
+    return label_map.get(pred_label, "Unknown")
+
+def predict_hate_speech(audio_path=None, text=None):
+    if audio_path:
+        transcription = transcribe(audio_path)
+        text_input = text if text else transcription
+    elif text:
+        text_input = text
     else:
-        prediction = extract_text_features(text_input)
-        return f"Predicted: {prediction}"
+        return "No input provided"
 
-st.title("Hate Speech Detector")
-uploaded_audio = st.file_uploader("Upload Audio File (.mp3, .wav, .ogg, .flac, .opus)", type=["mp3", "wav", "ogg", "flac", "opus"])
-text_input = st.text_input("Or enter text:")
+    prediction = extract_text_features(text_input)
+    return prediction
+
+st.title("Hate Speech Detector with Audio and Text")
+audio_file = st.file_uploader("Upload an audio file", type=["wav", "mp3", "flac", "ogg", "opus"])
+text_input = st.text_input("Optional text input")
 if st.button("Predict"):
-    result = predict(uploaded_audio, text_input)
-    st.success(result)
+    if audio_file is not None:
+        with open("temp_audio.wav", "wb") as f:
+            f.write(audio_file.read())
+        prediction = predict_hate_speech("temp_audio.wav", text_input)
+        st.success(prediction)
+    elif text_input:
+        prediction = predict_hate_speech(text=text_input)
+        st.success(prediction)
+    else:
+        st.warning("Please upload an audio file or enter text.")