Update app.py

Testing youtube transcrip TTS

app.py

@@ -1,10 +1,11 @@
+help me analyse this code, it is for a tts hugginface space
 import spaces
 import gradio as gr
 import edge_tts
 import asyncio
 import tempfile
 import os
-import re  # Import the regular expression module
+import re
 from pathlib import Path
 from pydub import AudioSegment
 
@@ -14,11 +15,11 @@ def get_silence(duration_ms=1000):
         duration=duration_ms,
         frame_rate=24000  # 24kHz sampling rate
     )
-    
+
     # Set audio parameters
     silent_audio = silent_audio.set_channels(1)  # Mono
     silent_audio = silent_audio.set_sample_width(4)  # 32-bit (4 bytes per sample)
-    
+
     with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
         # Export with specific bitrate and codec parameters
         silent_audio.export(
@@ -39,178 +40,131 @@ async def get_voices():
     voices = await edge_tts.list_voices()
     return {f"{v['ShortName']} - {v['Locale']} ({v['Gender']})": v['ShortName'] for v in voices}
 
-# Text-to-speech function for a single paragraph with SS handling
-async def paragraph_to_speech(text, voice, rate, pitch):
-    voice1 = "en-AU-WilliamNeural - en-AU (Male)"    
-    voice1F ="en-GB-SoniaNeural - en-GB (Female)"
-    voice2 = "en-GB-RyanNeural - en-GB (Male)"
-    voice2F = "en-US-JennyNeural - en-US (Female)"
-    voice3 ="en-US-BrianMultilingualNeural - en-US (Male)"  #good for reading    
-    voice3F = "en-HK-YanNeural - en-HK (Female)"
-    voice4 = "en-GB-ThomasNeural - en-GB (Male)"
-    voice4F ="en-US-EmmaNeural - en-US (Female)"      
-    voice5 = "en-GB-RyanNeural - en-GB (Male)" #Old Man
-    voice6 = "en-GB-MaisieNeural - en-GB (Female)"  #Child
-
-    if not text.strip():
-        return None, []  # Return None for audio path and empty list for silence
-
-    audio_segments = []
-    silence_durations = []
-    parts = re.split(r'(SS\d+\.?\d*)', text)  #this one separtate the SS## tag if any in the text.
-    for part in parts:          
-        if re.match(r'SS\d+\.?\d*', part):  #Check if there is Silence tag
-            # At the top of your file:
-            #SILENCE_PATH = Path(__file__).parent.absolute() / "Silence.mp3"
-            # At the top of your file (assuming you uploaded "Silence.mp3" to root)
-            #SILENCE_PATH = Path(__file__).parent.absolute() / "Silence.mp3"
-            # At the top of your file:
-            #SILENCE_PATH = Path(__file__).parent.absolute() / "static" / "intro.mp3"
-            #if SILENCE_PATH.exists():
-            #    audio_segments.append(str(SILENCE_PATH))
-            #    print(f"Silence.mp3 file found at {SILENCE_PATH} and is inserted")
-            #else:
-            silence_duration = float(part[2:]) * 1000  # Convert to milliseconds
-            print(f"Silence.mp3 file NOT FOUND")
-            silence_file_path = get_silence(silence_duration)  # Store the returned filename
-            audio_segments.append(silence_file_path)  # Use the stored filename
-        elif part.strip():
-            detect=0
-            processed_text = part
-            current_voice = voice
-            current_rate = rate
-            current_pitch = pitch
-            if part.startswith("1F"):
-                detect=1
-                current_voice = voice1F.split(" - ")[0]
-                current_pitch = 25
-            elif part.startswith("2F"):
-                detect=1
-                current_voice = voice2F.split(" - ")[0]
-            elif part.startswith("3F"):
-                detect=1
-                current_voice = voice3F.split(" - ")[0]
-            elif part.startswith("4F"):
-                detect=1
-                current_voice = voice4F.split(" - ")[0]                
-            elif part.startswith("1M"):
-                detect=1
-                current_voice = voice1.split(" - ")[0]
-            elif part.startswith("2M"):
-                detect=1
-                current_voice = voice2.split(" - ")[0]
-            elif part.startswith("3M"):
-                detect=1
-                current_voice = voice3.split(" - ")[0]
-            elif part.startswith("4M"):
-                detect=1
-                current_voice = voice4.split(" - ")[0]              
-            elif part.startswith("1O"):  # Old man voice
-                detect=1
-                current_voice = voice5.split(" - ")[0]
-                current_pitch = -20
-                current_rate = -10
-            elif part.startswith("1C"):  #Child voice
-                detect=1
-                current_voice = voice6.split(" - ")[0]                
-            else:
-                # Use selected voice, or fallback to default
-                #voice_short_name = (voice or default_voice).split(" - ")[0]
-                current_voice = (voice or default_voice).split(" - ")[0]
-                processed_text=part[:]
-                # Step 1: Use regex to find the first number, possibly negative, after a prefix (e.g., F-)
-            #match = re.search(r'[A-Za-z]\d+', part)  # Look for a letter followed by one or more digits
-            match = re.search(r'[A-Za-z]+\-?\d+', part)  # Look for a letter(s) followed by an optional '-' and digits
-            if match:
-                # Extract the prefix (e.g., '2F') and number (e.g., '-20')
-                prefix = ''.join([ch for ch in match.group() if ch.isalpha()])  # Extract letters (prefix)
-                number = int(''.join([ch for ch in match.group() if ch.isdigit() or ch == '-']))  # Extract digits (number)
-                current_pitch = number
-                # Step 2: Remove the found number from the string
-                new_text = re.sub(r'[A-Za-z]+\-?\d+', '', part, count=1).strip()  # Remove prefix and number (e.g., '2F-20')
-                #processed_text = new_text[2:]  #cut out the prefix like 1F, 3M etc
-                processed_text = new_text[len(prefix):]  # Dynamically remove the prefix part
-            else:
-                if detect:
-                   processed_text = part[2:]
-            rate_str = f"{current_rate:+d}%"
-            #if  part[2:4].isdigit():
-            #    processed_text = part[4:]
-            #    pitch = int(part[2:4])
-            pitch_str = f"{current_pitch:+d}Hz"
-            communicate = edge_tts.Communicate(processed_text, current_voice, rate=rate_str, pitch=pitch_str)
-            with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
-                tmp_path = tmp_file.name
-                await communicate.save(tmp_path)
-            audio_segments.append(tmp_path)
-        else:
-            audio_segments.append(None) # Empty string
-
-    return audio_segments, silence_durations
-
-# Main text-to-speech function that processes paragraphs and silence
-async def text_to_speech(text, voice, rate, pitch):
-    if not text.strip():
-        return None, gr.Warning("Please enter text to convert.")
+async def process_transcript_line(line, voice, rate, pitch):
+    """Processes a single transcript line to extract time, voice commands, and generate audio."""
+    match = re.match(r'(\d+):(\d+)(?:\.(\d+))?\s+(.*)', line)
+    if match:
+        minutes, seconds, milliseconds_str, text_with_commands = match.groups()
+        start_time_ms = int(minutes) * 60000 + int(seconds) * 1000 + (int(milliseconds_str) * 10 if milliseconds_str else 0)
+        if not text_with_commands.strip():
+            return start_time_ms, None
+
+        current_voice = voice
+        current_rate = rate
+        current_pitch = pitch
+        processed_text = text_with_commands
+
+        voice1 = "en-AU-WilliamNeural - en-AU (Male)"
+        voice1F ="en-GB-SoniaNeural - en-GB (Female)"
+        voice2 = "en-GB-RyanNeural - en-GB (Male)"
+        voice2F = "en-US-JennyNeural - en-US (Female)"
+        voice3 ="en-US-BrianMultilingualNeural - en-US (Male)"  #good for reading
+        voice3F = "en-HK-YanNeural - en-HK (Female)"
+        voice4 = "en-GB-ThomasNeural - en-GB (Male)"
+        voice4F ="en-US-EmmaNeural - en-US (Female)"
+        voice5 = "en-GB-RyanNeural - en-GB (Male)" #Old Man
+        voice6 = "en-GB-MaisieNeural - en-GB (Female)"  #Child
+
+        if text_with_commands.startswith("1F"):
+            current_voice = voice1F.split(" - ")[0]
+            current_pitch = 25
+            processed_text = text_with_commands[2:].strip()
+        elif text_with_commands.startswith("2F"):
+            current_voice = voice2F.split(" - ")[0]
+            processed_text = text_with_commands[2:].strip()
+        elif text_with_commands.startswith("3F"):
+            current_voice = voice3F.split(" - ")[0]
+            processed_text = text_with_commands[2:].strip()
+        elif text_with_commands.startswith("4F"):
+            current_voice = voice4F.split(" - ")[0]
+            processed_text = text_with_commands[2:].strip()
+        elif text_with_commands.startswith("1M"):
+            current_voice = voice1.split(" - ")[0]
+            processed_text = text_with_commands[2:].strip()
+        elif text_with_commands.startswith("2M"):
+            current_voice = voice2.split(" - ")[0]
+            processed_text = text_with_commands[2:].strip()
+        elif text_with_commands.startswith("3M"):
+            current_voice = voice3.split(" - ")[0]
+            processed_text = text_with_commands[2:].strip()
+        elif text_with_commands.startswith("4M"):
+            current_voice = voice4.split(" - ")[0]
+            processed_text = text_with_commands[2:].strip()
+        elif text_with_commands.startswith("1O"):  # Old man voice
+            current_voice = voice5.split(" - ")[0]
+            current_pitch = -20
+            current_rate = -10
+            processed_text = text_with_commands[2:].strip()
+        elif text_with_commands.startswith("1C"):  #Child voice
+            current_voice = voice6.split(" - ")[0]
+            processed_text = text_with_commands[2:].strip()
+
+        rate_str = f"{current_rate:+d}%"
+        pitch_str = f"{current_pitch:+d}Hz"
+        communicate = edge_tts.Communicate(processed_text, current_voice, rate=rate_str, pitch=pitch_str)
+        with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
+            audio_path = tmp_file.name
+            await communicate.save(audio_path)
+        return start_time_ms, audio_path
+    return None, None
+
+async def transcript_to_speech(transcript_text, voice, rate, pitch):
+    if not transcript_text.strip():
+        return None, gr.Warning("Please enter transcript text.")
     if not voice:
         return None, gr.Warning("Please select a voice.")
 
-    #paragraphs = [p.strip() for p in re.split(r'"', text) if p.strip()]
-    # Split the text using straight quotes (") and curly quotes (“ and ”)
-    paragraphs = [p.strip() for p in re.split(r'[“”"]', text) if p.strip()]
-    final_audio_segments = []
+    lines = transcript_text.strip().split('\n')
+    audio_segments_with_time = []
+    max_end_time_ms = 0
+
+    for line in lines:
+        start_time, audio_path = await process_transcript_line(line, voice, rate, pitch)
+        if start_time is not None and audio_path:
+            audio = AudioSegment.from_mp3(audio_path)
+            audio_segments_with_time.append({'start': start_time, 'audio': audio, 'path': audio_path})
+            max_end_time_ms = max(max_end_time_ms, start_time + len(audio))
+        elif audio_path:
+            os.remove(audio_path) # Clean up even if no timestamp
 
-    for paragraph in paragraphs:
-        audio_paths, silence_times = await paragraph_to_speech(paragraph, voice, rate, pitch)
-        if audio_paths:
-            for i, path in enumerate(audio_paths):
-                final_audio_segments.append(path)
-                if i < len(silence_times):
-                    final_audio_segments.append(silence_times[i])
+    if not audio_segments_with_time:
+        return None, "No valid transcript lines found."
 
-    if not any(isinstance(item, str) for item in final_audio_segments):
-        return None, None  # No actual audio generated
+    # Create initial silence audio
+    final_audio = AudioSegment.silent(duration=max_end_time_ms, frame_rate=24000)
 
-    if all(not isinstance(item, str) for item in final_audio_segments):
-        return None, "Only silence markers found."
+    for segment in audio_segments_with_time:
+        final_audio = final_audio.overlay(segment['audio'], position=segment['start'])
+        os.remove(segment['path']) # Clean up individual audio files
 
     combined_audio_path = tempfile.mktemp(suffix=".mp3")
-    with open(combined_audio_path, 'wb') as outfile:
-        for segment in final_audio_segments:
-            if isinstance(segment, str):
-                try:
-                    with open(segment, 'rb') as infile:
-                        outfile.write(infile.read())
-                    os.remove(segment)  # Clean up individual files
-                except FileNotFoundError:
-                    print(f"Warning: Audio file not found: {segment}")
+    final_audio.export(combined_audio_path, format="mp3")
     return combined_audio_path, None
 
-# Gradio interface function
 @spaces.GPU
-def tts_interface(text, voice, rate, pitch):
-    audio, warning = asyncio.run(text_to_speech(text, voice, rate, pitch))
+def tts_interface(transcript, voice, rate, pitch):
+    audio, warning = asyncio.run(transcript_to_speech(transcript, voice, rate, pitch))
     return audio, warning
 
-# Create Gradio application
-import gradio as gr
-
 async def create_demo():
     voices = await get_voices()
-    default_voice = "en-US-AndrewMultilingualNeural - en-US (Male)"  # 👈 Pick one of the available voices
+    default_voice = "en-US-AndrewMultilingualNeural - en-US (Male)"
     description = """
-    Default = <b>"en-US-AndrewMultilingualNeural - en-US (Male), 
-    other voices 1F:en-GB-SoniaNeural,    2F:en-US-JennyNeural,  3F:en-HK-YanNeural,  4F:en-US-EmmaNeural
-                 1M:en-AU-WilliamNeural,  2M:en-GB-RyanNeural,   3M:en-US-BrianMultilingualNeural,  4M:en-GB-ThomasNeural
-                 1C: en-GB-MaisieNeural (Childvoice), 1O = en-GB-RyanNeural (OldMan)"</b>
-    You can insert silence using the marker 'SS##' example "SS2.0" 
-    Enter your text, select a voice, and adjust the speech rate and pitch. Can also set like 1F-20  or 1M24.
+    Process YouTube transcript text with timestamps to generate synchronized audio.
+    Each line should be in the format: `minutes:seconds[.milliseconds] text`.
+    Voice prefixes (e.g., 1F, 1C) can be used at the beginning of a line to switch voices.
+    Example:
+    ```
+    0:00 This
+    0:14 is the story of little Red Riding Hood
+    0:38 1F Grandma isn’t feeling very well.
+    0:48 1C Yes, said Little Red Riding Hood.
+    ```
     """
-
     demo = gr.Interface(
         fn=tts_interface,
         inputs=[
-            gr.Textbox(label="Input Text", lines=5, placeholder="Separate paragraphs with two blank lines. Use 'SS[duration]' for silence."),
+            gr.Textbox(label="YouTube Transcript", lines=10, placeholder="0:00 This\n0:14 is the story...\n0:38 1F Grandma..."),
             gr.Dropdown(choices=[""] + list(voices.keys()), label="Select Voice", value=default_voice),
             gr.Slider(minimum=-50, maximum=50, value=0, label="Speech Rate Adjustment (%)", step=1),
             gr.Slider(minimum=-50, maximum=50, value=0, label="Pitch Adjustment (Hz)", step=1)
@@ -219,15 +173,13 @@ async def create_demo():
             gr.Audio(label="Generated Audio", type="filepath"),
             gr.Markdown(label="Warning", visible=False)
         ],
-        title="TTS using Edge Engine.. ENGLISH!",
+        title="TTS for YouTube Transcripts with Voice Switching",
         description=description,
-        article="Process text paragraph by paragraph for smoother output and insert silence markers.",
         analytics_enabled=False,
         allow_flagging=False
     )
     return demo
 
-# Run the application
 if __name__ == "__main__":
     demo = asyncio.run(create_demo())
     demo.launch()