Update app.py

app.py

@@ -6,6 +6,9 @@ import ffmpeg
 import nltk
 import re
 from deep_translator import MyMemoryTranslator
+import soundfile as sf
+from gradio_client_api import Client, handle_file
+
 
 # You only need to run this download command once
 nltk.download('punkt_tab')
@@ -54,6 +57,34 @@ def transcribe_audio(audio_path):
     """
     # Initialize the ASR pipeline from Hugging Face Transformers
     
+    THRESHOLDS = {
+    "very_slow":  80,
+    "slow":       110,
+    "normal":     150,
+    "fast":       200,
+    "very_fast":  float("inf")
+    }
+
+    def get_audio_duration(path: str) -> float:
+        """Return duration of audio file in seconds."""
+        with sf.SoundFile(path) as f:
+            return len(f) / f.samplerate
+
+    def compute_wpm(transcript: str, duration_s: float) -> float:
+        """Compute words per minute."""
+        if not transcript or duration_s == 0:
+            return 0.0
+        words = transcript.strip().split()
+        return len(words) / (duration_s / 60.0)
+
+    def categorize_wpm(wpm: float) -> str:
+        """Map a WPM value to one of the pace categories."""
+        for label, threshold in THRESHOLDS.items():
+            if wpm < threshold:
+                return label
+        return "unknown"
+
+    
     transcriber = pipeline(
         "automatic-speech-recognition",
         model="openai/whisper-large-v3-turbo",
@@ -66,16 +97,19 @@ def transcribe_audio(audio_path):
         return "No audio file provided. Please upload or record an audio file."
     
     print(f"Transcribing audio file: {audio_path}")
-    try:
-        # The pipeline handles all the complex steps of loading and processing the audio
-        result = transcriber(audio_path)
-        # The result is a dictionary, and we need the 'text' key
-        transcription = result["text"]
-        print(f"Transcription successful: {transcription}")
-        return transcription
-    except Exception as e:
-        print(f"An error occurred during transcription: {e}")
-        return f"Sorry, an error occurred. Please try again. Details: {str(e)}"
+    
+    # The pipeline handles all the complex steps of loading and processing the audio
+    result = transcriber(audio_path)
+    # The result is a dictionary, and we need the 'text' key
+    transcription = result["text"]
+    print(f"Transcription successful: {transcription}")
+
+    duration_s = get_audio_duration(audio_path)
+    wpm = compute_wpm(transcribed_text, duration_s)
+    pace = categorize_wpm(wpm)
+    print(f"    > Pace detected: {pace.upper()} ({wpm:.1f} WPM)")
+    
+    return transcription, pace
 
 def lang_select(target_lang):
     LANGUAGE_NAME_TO_CODE = {
@@ -128,17 +162,38 @@ def translate_local(text_to_translate, target_lang='ta-IN', device=None):
 
     return translated_text
 
+def synthesize_speech(synth_text, target_lang, gender="Male", pace="normal", output_path="temp_audio_synthesized.wav", device=None):
+
+    ref_audio_path = str('reference/'+target_lang.upper()+'/'+gender.upper()+'_'+pace.upper()+'.wav')
+    ref_text_path = str('reference/'+target_lang.upper()+'/'+gender.upper()+'_'+pace.upper()+'.txt')
+
+    if os.path.exists(ref_audio_path) and os.path.exists(ref_text_path):
+        ref_audio_path = ref_audio_path
+        with open(ref_text_path, encoding='utf-8') as f:
+            ref_text = f.read()
+
+    client = Client("ai4bharat/IndicF5")
+    result = client.predict(
+            text=synth_text,
+            ref_audio=handle_file(ref_audio_path),
+            ref_text=ref_text,
+            api_name="/synthesize_speech"
+    )
+    print("synthesize_speech saved to ", result)
+    return result
+
 def main_run(video_path,target_lang):
     original_audio_file = extract_audio_from_video(video_path)
-    original_text = transcribe_audio(original_audio_file)
+    original_text , pace = transcribe_audio(original_audio_file)
     translated_text = translate_local(original_text,target_lang)
-    (original_text,target_lang)
-    return translated_text
+    translated_audio = synthesize_speech(translated_text, target_lang, "Male", pace)
+    return translated_audio
 
 def audio_pipeline_run(audio_path,target_lang):
-    original_text = transcribe_audio(audio_path)
+    original_text , pace = transcribe_audio(audio_path)
     translated_text = translate_local(original_text,target_lang)
-    return translated_text
+    translated_audio = synthesize_speech(translated_text, target_lang, "Male", pace)
+    return translated_audio
 
 
 # --- Gradio Interface Definition ---
@@ -176,7 +231,7 @@ with gr.Blocks(title="Audio/Video Translation Toolkit") as app_interface:
                     submit_btn_vid = gr.Button("Translate Video", variant="primary")
 
                 with gr.Column():
-                    video_out = gr.Textbox(label="Translation")
+                    video_out = gr.Audio(label="Output Audio", interactive=False)
 
                 '''
                 with gr.Column():
@@ -191,15 +246,10 @@ with gr.Blocks(title="Audio/Video Translation Toolkit") as app_interface:
                     audio_in_pipe = gr.Audio(type="filepath", label="Input Audio")
                     lang_radio_aud = gr.Radio(choices=["Tamil", "Telugu", "Hindi"], label="Target Language", value="Tamil")
                     submit_btn_aud = gr.Button("Translate Audio", variant="primary")
-
-                with gr.Column():
-                    audio_out_pipe = gr.Textbox(label="Translation")
                 
-                '''
                 with gr.Column():
                     audio_out_pipe = gr.Audio(label="Output Audio", interactive=False)
-                '''
-                
+                                
             submit_btn_aud.click(fn=audio_pipeline_run, inputs=[audio_in_pipe, lang_radio_aud], outputs=[audio_out_pipe])
 
 # --- Launch the App ---