Update app.py

app.py

@@ -2,7 +2,6 @@ import os
 import streamlit as st
 import tempfile
 import torch
-import torchaudio
 import transformers
 from transformers import pipeline, AutoModelForSequenceClassification, AutoTokenizer
 import plotly.express as px
@@ -15,6 +14,15 @@ import asyncio
 from concurrent.futures import ThreadPoolExecutor
 import streamlit.components.v1 as components
 
+# Try importing torchaudio, fallback to pydub
+try:
+    import torchaudio
+    USE_TORCHAUDIO = True
+except ImportError:
+    from pydub import AudioSegment
+    USE_TORCHAUDIO = False
+    st.warning("torchaudio not found. Using pydub (slower). Install torchaudio: pip install torchaudio")
+
 # Suppress warnings
 logging.getLogger("torch").setLevel(logging.ERROR)
 logging.getLogger("transformers").setLevel(logging.ERROR)
@@ -33,21 +41,25 @@ st.markdown("Fast, accurate detection of emotions, sentiment, and sarcasm from v
 # Global model cache
 @st.cache_resource
 def load_models():
-    whisper_model = whisper.load_model("base")
-    
-    emotion_tokenizer = AutoTokenizer.from_pretrained("bhadresh-savani/distilbert-base-uncased-emotion")
-    emotion_model = AutoModelForSequenceClassification.from_pretrained("bhadresh-savani/distilbert-base-uncased-emotion")
-    emotion_model = emotion_model.to(device).half()
-    emotion_classifier = pipeline("text-classification", model=emotion_model, tokenizer=emotion_tokenizer,
-                                 top_k=None, device=0 if torch.cuda.is_available() else -1)
+    try:
+        whisper_model = whisper.load_model("base")
+        
+        emotion_tokenizer = AutoTokenizer.from_pretrained("bhadresh-savani/distilbert-base-uncased-emotion")
+        emotion_model = AutoModelForSequenceClassification.from_pretrained("bhadresh-savani/distilbert-base-uncased-emotion")
+        emotion_model = emotion_model.to(device).half()
+        emotion_classifier = pipeline("text-classification", model=emotion_model, tokenizer=emotion_tokenizer,
+                                     top_k=None, device=0 if torch.cuda.is_available() else -1)
 
-    sarcasm_tokenizer = AutoTokenizer.from_pretrained("cardiffnlp/twitter-roberta-base-irony")
-    sarcasm_model = AutoModelForSequenceClassification.from_pretrained("cardiffnlp/twitter-roberta-base-irony")
-    sarcasm_model = sarcasm_model.to(device).half()
-    sarcasm_classifier = pipeline("text-classification", model=sarcasm_model, tokenizer=sarcasm_tokenizer,
-                                 device=0 if torch.cuda.is_available() else -1)
-    
-    return whisper_model, emotion_classifier, sarcasm_classifier
+        sarcasm_tokenizer = AutoTokenizer.from_pretrained("cardiffnlp/twitter-roberta-base-irony")
+        sarcasm_model = AutoModelForSequenceClassification.from_pretrained("cardiffnlp/twitter-roberta-base-irony")
+        sarcasm_model = sarcasm_model.to(device).half()
+        sarcasm_classifier = pipeline("text-classification", model=sarcasm_model, tokenizer=sarcasm_tokenizer,
+                                     device=0 if torch.cuda.is_available() else -1)
+        
+        return whisper_model, emotion_classifier, sarcasm_classifier
+    except Exception as e:
+        st.error(f"Failed to load models: {str(e)}")
+        raise
 
 whisper_model, emotion_classifier, sarcasm_classifier = load_models()
 
@@ -90,29 +102,45 @@ async def perform_sarcasm_detection(text):
 # Audio validation
 def validate_audio(audio_path):
     try:
-        waveform, sample_rate = torchaudio.load(audio_path)
-        if waveform.abs().max() < 0.01:
-            st.warning("Audio volume too low.")
-            return False
-        if waveform.shape[1] / sample_rate < 1:
-            st.warning("Audio too short.")
-            return False
+        if USE_TORCHAUDIO:
+            waveform, sample_rate = torchaudio.load(audio_path)
+            if waveform.abs().max() < 0.01:
+                st.warning("Audio volume too low.")
+                return False
+            if waveform.shape[1] / sample_rate < 1:
+                st.warning("Audio too short.")
+                return False
+        else:
+            sound = AudioSegment.from_file(audio_path)
+            if sound.dBFS < -55:
+                st.warning("Audio volume too low.")
+                return False
+            if len(sound) < 1000:
+                st.warning("Audio too short.")
+                return False
         return True
-    except:
-        st.error("Invalid audio file.")
+    except Exception as e:
+        st.error(f"Invalid audio file: {str(e)}")
         return False
 
 # Audio transcription
 @st.cache_data
 def transcribe_audio(audio_path):
     try:
-        waveform, sample_rate = torchaudio.load(audio_path)
-        if sample_rate != 16000:
-            resampler = torchaudio.transforms.Resample(sample_rate, 16000)
-            waveform = resampler(waveform)
-        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_file:
-            torchaudio.save(temp_file.name, waveform, 16000)
-            result = whisper_model.transcribe(temp_file.name, language="en")
+        if USE_TORCHAUDIO:
+            waveform, sample_rate = torchaudio.load(audio_path)
+            if sample_rate != 16000:
+                resampler = torchaudio.transforms.Resample(sample_rate, 16000)
+                waveformვ: waveform = resampler(waveform)
+            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_file:
+                torchaudio.save(temp_file.name, waveform, 16000)
+                result = whisper_model.transcribe(temp_file.name, language="en")
+        else:
+            sound = AudioSegment.from_file(audio_path)
+            sound = sound.set_frame_rate(16000).set_channels(1)
+            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_file:
+                sound.export(temp_file.name, format="wav")
+                result = whisper_model.transcribe(temp_file.name, language="en")
         os.remove(temp_file.name)
         return result["text"].strip()
     except Exception as e:
@@ -283,7 +311,8 @@ def main():
                         display_analysis_results(text)
                     else:
                         st.error("Transcription failed.")
-                    os.remove(temp_path)
+                    if os.path.exists(temp_path):
+                        os.remove(temp_path)
                 progress.empty()
     
     with tab2:
@@ -300,7 +329,8 @@ def main():
                     display_analysis_results(text)
                 else:
                     st.error("Transcription failed.")
-                os.remove(temp_path)
+                if os.path.exists(temp_path):
+                    os.remove(temp_path)
             progress.empty()
     
     with tab3: