Update app.py

app.py

@@ -1,9 +1,8 @@
-import os
+import streamlit as st
 import torch
 import torchaudio
 import numpy as np
 import re
-import streamlit as st
 from difflib import SequenceMatcher
 from transformers import pipeline
 
@@ -19,17 +18,18 @@ pipe = pipeline(
     chunk_length_s=30,
     device=device,
     generate_kwargs={
-        "no_repeat_ngram_size": 3,
-        "repetition_penalty": 1.3,
-        "temperature": 0.7,
-        "top_p": 0.9,
-        "top_k": 50
+        "no_repeat_ngram_size": 4,
+        "repetition_penalty": 1.15,
+        "temperature": 0.5,
+        "top_p": 0.97,
+        "top_k": 40,
+        "max_new_tokens": 300,
+        "do_sample": True
     }
 )
 pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language=language, task="transcribe")
 
-# Load quality rating model
-rating_pipe = pipeline("text-classification", model="Leo0129/CustomModel-multilingual-sentiment-analysis", device=device)
+rating_pipe = pipeline("text-classification", model="MonkeyDLLLLLLuffy/CustomModel-multilingual-sentiment-analysis", device=device)
 
 def is_similar(a, b, threshold=0.8):
     return SequenceMatcher(None, a, b).ratio() > threshold
@@ -47,57 +47,41 @@ def remove_punctuation(text):
 
 def transcribe_audio(audio_path):
     waveform, sample_rate = torchaudio.load(audio_path)
-    
     if waveform.shape[0] > 1:
         waveform = torch.mean(waveform, dim=0, keepdim=True)
-    
     waveform = waveform.squeeze(0).numpy()
     duration = waveform.shape[0] / sample_rate
-    
     if duration > 60:
         chunk_size = sample_rate * 55
         step_size = sample_rate * 50
         results = []
-        
         for start in range(0, waveform.shape[0], step_size):
             chunk = waveform[start:start + chunk_size]
             if chunk.shape[0] == 0:
                 break
             transcript = pipe({"sampling_rate": sample_rate, "raw": chunk})["text"]
             results.append(remove_punctuation(transcript))
-        
         return remove_punctuation(remove_repeated_phrases(" ".join(results)))
-    
     return remove_punctuation(remove_repeated_phrases(pipe({"sampling_rate": sample_rate, "raw": waveform})["text"]))
 
 def rate_quality(text):
     chunks = [text[i:i+512] for i in range(0, len(text), 512)]
     results = rating_pipe(chunks, batch_size=4)
-    
     label_map = {"Very Negative": "Very Poor", "Negative": "Poor", "Neutral": "Neutral", "Positive": "Good", "Very Positive": "Very Good"}
     processed_results = [label_map.get(res["label"], "Unknown") for res in results]
-    
     return max(set(processed_results), key=processed_results.count)
 
 # Streamlit UI
-st.title("Audio Transcription and Quality Rating")
-
+st.title("Audio Transcription & Quality Rating")
 uploaded_file = st.file_uploader("Upload an audio file", type=["wav", "mp3", "flac"])
-
-if uploaded_file is not None:
-    st.audio(uploaded_file, format="audio/wav")
-    
-    temp_audio_path = "temp_audio.wav"
-    with open(temp_audio_path, "wb") as f:
+if uploaded_file:
+    st.audio(uploaded_file, format='audio/wav')
+    with open("temp_audio.wav", "wb") as f:
         f.write(uploaded_file.read())
-    
     st.write("Processing audio...")
-    transcript = transcribe_audio(temp_audio_path)
+    transcript = transcribe_audio("temp_audio.wav")
     st.subheader("Transcript")
     st.write(transcript)
-    
     quality_rating = rate_quality(transcript)
     st.subheader("Quality Rating")
     st.write(quality_rating)
-    
-    os.remove(temp_audio_path)