Update app.py

app.py

@@ -1,7 +1,6 @@
 from flask import Flask, render_template, request, jsonify
 import os
 import torch
-import speech_recognition as sr
 from transformers import pipeline
 from gtts import gTTS
 import re
@@ -11,7 +10,7 @@ from waitress import serve
 
 app = Flask(__name__)
 
-# Load Whisper Model for Accurate Speech-to-Text
+# Load Whisper Model (Use whisper-small for better performance)
 device = "cuda" if torch.cuda.is_available() else "cpu"
 asr_model = pipeline("automatic-speech-recognition", model="openai/whisper-small", device=0 if device == "cuda" else -1)
 
@@ -31,7 +30,7 @@ prompts = {
 for key, text in prompts.items():
     generate_audio_prompt(text, f"{key}.mp3")
 
-# Symbol mapping for better recognition
+# Symbol mapping for proper recognition
 SYMBOL_MAPPING = {
     "at the rate": "@",
     "at": "@",
@@ -46,22 +45,16 @@ SYMBOL_MAPPING = {
 
 # Function to clean and format transcribed text properly
 def clean_transcription(text):
-    text = text.lower()
-    text = re.sub(r"\s+", " ", text).strip()  # Remove extra spaces
+    text = text.lower().strip()
     for word, symbol in SYMBOL_MAPPING.items():
         text = text.replace(word, symbol)
     return text.capitalize()
 
-# Function to detect speech duration and avoid cutting words
-def trim_silence(audio_path):
+# Function to check if the audio contains actual speech
+def is_silent_audio(audio_path):
     audio = AudioSegment.from_wav(audio_path)
     nonsilent_parts = detect_nonsilent(audio, min_silence_len=700, silence_thresh=audio.dBFS-16)
-    
-    if nonsilent_parts:
-        start_trim = nonsilent_parts[0][0]
-        end_trim = nonsilent_parts[-1][1]
-        trimmed_audio = audio[start_trim:end_trim]
-        trimmed_audio.export(audio_path, format="wav")  # Save trimmed audio
+    return len(nonsilent_parts) == 0  # Returns True if silence detected
 
 @app.route("/")
 def index():
@@ -77,7 +70,9 @@ def transcribe():
     audio_file.save(audio_path)
 
     try:
-        trim_silence(audio_path)  # Trim silence before processing
+        # Check if audio contains valid speech
+        if is_silent_audio(audio_path):
+            return jsonify({"error": "No speech detected. Please try again."}), 400
         
         # Force Whisper to transcribe only in English
         result = asr_model(audio_path, generate_kwargs={"language": "en"})
@@ -85,8 +80,8 @@ def transcribe():
         
         return jsonify({"text": transcribed_text})
     except Exception as e:
-        return jsonify({"error": str(e)}), 500
+        return jsonify({"error": f"Speech recognition error: {str(e)}"}), 500
 
-# Use Waitress for Production Server
+# Run Waitress Production Server
 if __name__ == "__main__":
     serve(app, host="0.0.0.0", port=7860)