Update app.py

app.py

@@ -2,8 +2,7 @@ import torch
 from flask import Flask, render_template, request, jsonify
 import os
 import re
-import ffmpeg
-from transformers import pipeline  # ✅ Using correct Whisper ASR pipeline
+from transformers import pipeline
 from gtts import gTTS
 from pydub import AudioSegment
 from pydub.silence import detect_nonsilent
@@ -11,11 +10,11 @@ from waitress import serve
 
 app = Flask(__name__)
 
-# ✅ Load Whisper ASR Model correctly
+# ✅ Load Whisper ASR Model Correctly with Language Specification
 device = "cuda" if torch.cuda.is_available() else "cpu"
 asr_model = pipeline("automatic-speech-recognition", model="openai/whisper-large-v3", device=0 if device == "cuda" else -1)
 
-# Function to generate audio prompts
+# Function to generate voice prompts
 def generate_audio_prompt(text, filename):
     tts = gTTS(text=text, lang="en")
     tts.save(os.path.join("static", filename))
@@ -31,44 +30,26 @@ prompts = {
 for key, text in prompts.items():
     generate_audio_prompt(text, f"{key}.mp3")
 
-# Symbol mapping for proper recognition
-SYMBOL_MAPPING = {
-    "at the rate": "@",
-    "at": "@",
-    "dot": ".",
-    "underscore": "_",
-    "hash": "#",
-    "plus": "+",
-    "dash": "-",
-    "comma": ",",
-    "space": " "
-}
-
-# Function to convert audio to WAV format
+# ✅ Ensure Proper Audio Format (16kHz, Mono)
 def convert_to_wav(input_path, output_path):
     try:
         audio = AudioSegment.from_file(input_path)
+        audio = audio.set_frame_rate(16000).set_channels(1)  # ✅ Convert to 16kHz, mono
         audio.export(output_path, format="wav")
     except Exception as e:
         raise Exception(f"Audio conversion failed: {str(e)}")
 
-# Function to clean transcribed text
-def clean_transcription(text):
-    text = text.lower().strip()
-    ignore_phrases = ["my name is", "this is", "i am", "it's", "name"]
-    for phrase in ignore_phrases:
-        text = text.replace(phrase, "").strip()
-    
-    for word, symbol in SYMBOL_MAPPING.items():
-        text = text.replace(word, symbol)
-    
-    return text.capitalize()
-
-# Function to check if audio contains actual speech
+# ✅ Check for Silence
 def is_silent_audio(audio_path):
     audio = AudioSegment.from_wav(audio_path)
     nonsilent_parts = detect_nonsilent(audio, min_silence_len=500, silence_thresh=audio.dBFS-16)
-    return len(nonsilent_parts) == 0  # Returns True if silence detected
+    return len(nonsilent_parts) == 0
+
+# ✅ Clean Transcription Text
+def clean_transcription(text):
+    text = text.strip()
+    text = re.sub(r"[-.]", "", text)  # ✅ Remove unwanted characters
+    return text.capitalize()
 
 @app.route("/")
 def index():
@@ -85,21 +66,21 @@ def transcribe():
     audio_file.save(input_audio_path)
 
     try:
-        # Convert to WAV
+        # ✅ Convert audio to proper format
         convert_to_wav(input_audio_path, output_audio_path)
 
-        # Check for silence
+        # ✅ Check for silent audio
         if is_silent_audio(output_audio_path):
             return jsonify({"error": "No speech detected. Please try again."}), 400
-        
-        # ✅ Use Whisper ASR model for transcription
+
+        # ✅ Transcribe Using Whisper ASR
         result = asr_model(output_audio_path, generate_kwargs={"language": "en"})
         transcribed_text = clean_transcription(result["text"])
-        
+
         return jsonify({"text": transcribed_text})
     except Exception as e:
         return jsonify({"error": f"Speech recognition error: {str(e)}"}), 500
 
-# Start Waitress Production Server
+# ✅ Start Production Server
 if __name__ == "__main__":
     serve(app, host="0.0.0.0", port=7860)