Update app.py

app.py

@@ -9,114 +9,82 @@ from pathlib import Path
 from pydub import AudioSegment
 
 def get_silence(duration_ms=1000):
-    # Create silent audio segment with specified parameters
-    silent_audio = AudioSegment.silent(
-        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(
-            tmp_file.name,
-            format="mp3",
-            bitrate="48k",
-            parameters=[
-                "-ac", "1",  # Mono
-                "-ar", "24000",  # Sample rate
-                "-sample_fmt", "s32",  # 32-bit samples
-                "-codec:a", "libmp3lame"  # MP3 codec
-            ]
-        )
-        return tmp_file.name
+    # ... (get_silence function remains the same)
 
 # Get all available voices
 async def get_voices():
-    voices = await edge_tts.list_voices()
-    return {f"{v['ShortName']} - {v['Locale']} ({v['Gender']})": v['ShortName'] for v in voices}
-
-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_full = voice  # Store the full voice string from the dropdown
-        current_voice_short = current_voice_full.split(" - ")[0] if current_voice_full else ""
-        current_rate = rate
-        current_pitch = pitch
-        processed_text = text_with_commands
-
-        voice1_full = "en-AU-WilliamNeural - en-AU (Male)"
-        voice1_short = voice1_full.split(" - ")[0]
-        voice1F_full ="en-GB-SoniaNeural - en-GB (Female)"
-        voice1F_short = voice1F_full.split(" - ")[0]
-        voice2_full = "en-GB-RyanNeural - en-GB (Male)"
-        voice2_short = voice2_full.split(" - ")[0]
-        voice2F_full = "en-US-JennyNeural - en-US (Female)"
-        voice2F_short = voice2F_full.split(" - ")[0]
-        voice3_full ="en-US-BrianMultilingualNeural - en-US (Male)"  #good for reading
-        voice3_short = voice3_full.split(" - ")[0]
-        voice3F_full = "en-HK-YanNeural - en-HK (Female)"
-        voice3F_short = voice3F_full.split(" - ")[0]
-        voice4_full = "en-GB-ThomasNeural - en-GB (Male)"
-        voice4_short = voice4_full.split(" - ")[0]
-        voice4F_full ="en-US-EmmaNeural - en-US (Female)"
-        voice4F_short = voice4F_full.split(" - ")[0]
-        voice5_full = "en-GB-RyanNeural - en-GB (Male)" #Old Man
-        voice5_short = voice5_full.split(" - ")[0]
-        voice6_full = "en-GB-MaisieNeural - en-GB (Female)"  #Child
-        voice6_short = voice6_full.split(" - ")[0]
-
-        if text_with_commands.startswith("1F"):
-            current_voice_short = voice1F_short
-            current_pitch = 25
-            processed_text = text_with_commands[2:].strip()
-        elif text_with_commands.startswith("2F"):
-            current_voice_short = voice2F_short
-            processed_text = text_with_commands[2:].strip()
-        elif text_with_commands.startswith("3F"):
-            current_voice_short = voice3F_short
-            processed_text = text_with_commands[2:].strip()
-        elif text_with_commands.startswith("4F"):
-            current_voice_short = voice4F_short
-            processed_text = text_with_commands[2:].strip()
-        elif text_with_commands.startswith("1M"):
-            current_voice_short = voice1_short
-            processed_text = text_with_commands[2:].strip()
-        elif text_with_commands.startswith("2M"):
-            current_voice_short = voice2_short
-            processed_text = text_with_commands[2:].strip()
-        elif text_with_commands.startswith("3M"):
-            current_voice_short = voice3_short
-            processed_text = text_with_commands[2:].strip()
-        elif text_with_commands.startswith("4M"):
-            current_voice_short = voice4_short
-            processed_text = text_with_commands[2:].strip()
-        elif text_with_commands.startswith("1O"):  # Old man voice
-            current_voice_short = voice5_short
-            current_pitch = -20
-            current_rate = -10
-            processed_text = text_with_commands[2:].strip()
-        elif text_with_commands.startswith("1C"):  #Child voice
-            current_voice_short = voice6_short
-            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_short, 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
+    # ... (get_voices function remains the same)
+
+async def text_to_speech_segment(text_segment, voice, rate, pitch):
+    """Processes a single text segment for voice commands and generates audio."""
+    current_voice_full = voice
+    current_voice_short = current_voice_full.split(" - ")[0] if current_voice_full else ""
+    current_rate = rate
+    current_pitch = pitch
+    processed_text = text_segment
+
+    voice1_full = "en-AU-WilliamNeural - en-AU (Male)"
+    voice1_short = voice1_full.split(" - ")[0]
+    voice1F_full ="en-GB-SoniaNeural - en-GB (Female)"
+    voice1F_short = voice1F_full.split(" - ")[0]
+    voice2_full = "en-GB-RyanNeural - en-GB (Male)"
+    voice2_short = voice2_full.split(" - ")[0]
+    voice2F_full = "en-US-JennyNeural - en-US (Female)"
+    voice2F_short = voice2F_full.split(" - ")[0]
+    voice3_full ="en-US-BrianMultilingualNeural - en-US (Male)"  #good for reading
+    voice3_short = voice3_full.split(" - ")[0]
+    voice3F_full = "en-HK-YanNeural - en-HK (Female)"
+    voice3F_short = voice3F_full.split(" - ")[0]
+    voice4_full = "en-GB-ThomasNeural - en-GB (Male)"
+    voice4_short = voice4_full.split(" - ")[0]
+    voice4F_full ="en-US-EmmaNeural - en-US (Female)"
+    voice4F_short = voice4F_full.split(" - ")[0]
+    voice5_full = "en-GB-RyanNeural - en-GB (Male)" #Old Man
+    voice5_short = voice5_full.split(" - ")[0]
+    voice6_full = "en-GB-MaisieNeural - en-GB (Female)"  #Child
+    voice6_short = voice6_full.split(" - ")[0]
+
+    if text_segment.startswith("1F"):
+        current_voice_short = voice1F_short
+        current_pitch = 25
+        processed_text = text_segment[2:].strip()
+    elif text_segment.startswith("2F"):
+        current_voice_short = voice2F_short
+        processed_text = text_segment[2:].strip()
+    elif text_segment.startswith("3F"):
+        current_voice_short = voice3F_short
+        processed_text = text_segment[2:].strip()
+    elif text_segment.startswith("4F"):
+        current_voice_short = voice4F_short
+        processed_text = text_segment[2:].strip()
+    elif text_segment.startswith("1M"):
+        current_voice_short = voice1_short
+        processed_text = text_segment[2:].strip()
+    elif text_segment.startswith("2M"):
+        current_voice_short = voice2_short
+        processed_text = text_segment[2:].strip()
+    elif text_segment.startswith("3M"):
+        current_voice_short = voice3_short
+        processed_text = text_segment[2:].strip()
+    elif text_segment.startswith("4M"):
+        current_voice_short = voice4_short
+        processed_text = text_segment[2:].strip()
+    elif text_segment.startswith("1O"):  # Old man voice
+        current_voice_short = voice5_short
+        current_pitch = -20
+        current_rate = -10
+        processed_text = text_segment[2:].strip()
+    elif text_segment.startswith("1C"):  #Child voice
+        current_voice_short = voice6_short
+        processed_text = text_segment[2:].strip()
+
+    rate_str = f"{current_rate:+d}%"
+    pitch_str = f"{current_pitch:+d}Hz"
+    communicate = edge_tts.Communicate(processed_text, current_voice_short, 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 audio_path
 
 async def transcript_to_speech(transcript_text, voice, rate, pitch):
     if not transcript_text.strip():
@@ -124,32 +92,67 @@ async def transcript_to_speech(transcript_text, voice, rate, pitch):
     if not voice:
         return None, gr.Warning("Please select a voice.")
 
-    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
-
-    if not audio_segments_with_time:
-        return None, "No valid transcript lines found."
-
-    # Create initial silence audio
-    final_audio = AudioSegment.silent(duration=max_end_time_ms, frame_rate=24000)
-
-    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")
-    final_audio.export(combined_audio_path, format="mp3")
-    return combined_audio_path, None
+    segments = re.split(r'[“”"]', transcript_text)
+    audio_paths = []
+
+    for segment in segments:
+        segment = segment.strip()
+        if segment:
+            # Check if the segment starts with a timestamp
+            timestamp_match = re.match(r'(\d+):(\d+)(?:\.(\d+))?\s+(.*)', segment)
+            if timestamp_match:
+                minutes, seconds, milliseconds_str, text_with_commands = timestamp_match.groups()
+                start_time_ms = int(minutes) * 60000 + int(seconds) * 1000 + (int(milliseconds_str) * 10 if milliseconds_str else 0)
+                audio_path = await text_to_speech_segment(text_with_commands, voice, rate, pitch)
+                audio_paths.append({'start': start_time_ms, 'path': audio_path})
+            else:
+                # Process segments without timestamps (for voice switching)
+                audio_path = await text_to_speech_segment(segment, voice, rate, pitch)
+                if audio_path:
+                    audio_paths.append({'start': None, 'path': audio_path}) # No specific start time
+
+    if not audio_paths:
+        return None, "No audio segments generated."
+
+    # Handle combining audio with timestamps
+    timed_segments = [item for item in audio_paths if item['start'] is not None]
+    non_timed_segments = [item for item in audio_paths if item['start'] is None and item['path']]
+
+    if timed_segments:
+        max_end_time_ms = 0
+        processed_timed_segments = []
+        for item in timed_segments:
+            audio = AudioSegment.from_mp3(item['path'])
+            processed_timed_segments.append({'start': item['start'], 'audio': audio, 'path': item['path']})
+            max_end_time_ms = max(max_end_time_ms, item['start'] + len(audio))
+
+        final_audio = AudioSegment.silent(duration=max_end_time_ms, frame_rate=24000)
+        for segment in processed_timed_segments:
+            final_audio = final_audio.overlay(segment['audio'], position=segment['start'])
+            os.remove(segment['path'])
+
+        # Append non-timed segments sequentially
+        for item in non_timed_segments:
+            audio = AudioSegment.from_mp3(item['path'])
+            final_audio += audio
+            os.remove(item['path'])
+
+        combined_audio_path = tempfile.mktemp(suffix=".mp3")
+        final_audio.export(combined_audio_path, format="mp3")
+        return combined_audio_path, None
+
+    elif non_timed_segments:
+        # Combine non-timed segments sequentially if no timestamps are found
+        combined_audio = AudioSegment.empty()
+        for item in non_timed_segments:
+            audio = AudioSegment.from_mp3(item['path'])
+            combined_audio += audio
+            os.remove(item['path'])
+        combined_audio_path = tempfile.mktemp(suffix=".mp3")
+        combined_audio.export(combined_audio_path, format="mp3")
+        return combined_audio_path, None
+
+    return None, "No processable audio segments found."
 
 @spaces.GPU
 def tts_interface(transcript, voice, rate, pitch):
@@ -160,21 +163,22 @@ async def create_demo():
     voices = await get_voices()
     default_voice = "en-US-AndrewMultilingualNeural - en-US (Male)"
     description = """
-    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.
+    Process text, handling both timestamped transcripts and voice switching using quote marks and prefixes.
+    Separate segments by quote marks ("). For timestamped segments, use the format: `minutes:seconds[.milliseconds] text`.
+    Voice prefixes (e.g., 1F, 1C) can be used at the beginning of a quoted segment 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.
+    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."
+    "and then the default voice continues"
     ```
     """
     demo = gr.Interface(
         fn=tts_interface,
         inputs=[
-            gr.Textbox(label="YouTube Transcript", lines=10, placeholder="0:00 This\n0:14 is the story...\n0:38 1F Grandma..."),
+            gr.Textbox(label="Input Text / Transcript", lines=10, placeholder='0:00 "This"\n"0:14 is the story..."\n"1F Hello"'),
             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)
@@ -183,7 +187,7 @@ async def create_demo():
             gr.Audio(label="Generated Audio", type="filepath"),
             gr.Markdown(label="Warning", visible=False)
         ],
-        title="TTS for YouTube Transcripts with Voice Switching",
+        title="Combined TTS: Timestamps and Voice Switching",
         description=description,
         analytics_enabled=False,
         allow_flagging=False