Update app.py

app.py

@@ -39,13 +39,13 @@ 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 text_to_speech_segment(text_segment, voice, rate, pitch):
-    """Processes a single text segment for voice commands and generates audio."""
-    current_voice_full = voice
+async def generate_audio_with_voice_prefix(text_segment, default_voice, rate, pitch):
+    """Generates audio for a text segment, handling voice prefixes."""
+    current_voice_full = default_voice
     current_voice_short = current_voice_full.split(" - ")[0] if current_voice_full else ""
     current_rate = rate
     current_pitch = pitch
-    processed_text = text_segment
+    processed_text = text_segment.strip()
 
     voice1_full = "en-AU-WilliamNeural - en-AU (Male)"
     voice1_short = voice1_full.split(" - ")[0]
@@ -68,47 +68,75 @@ async def text_to_speech_segment(text_segment, voice, rate, pitch):
     voice6_full = "en-GB-MaisieNeural - en-GB (Female)"  #Child
     voice6_short = voice6_full.split(" - ")[0]
 
-    if text_segment.startswith("1F"):
+    if processed_text.startswith("1F"):
         current_voice_short = voice1F_short
         current_pitch = 25
-        processed_text = text_segment[2:].strip()
-    elif text_segment.startswith("2F"):
+        processed_text = processed_text[2:].strip()
+    elif processed_text.startswith("2F"):
         current_voice_short = voice2F_short
-        processed_text = text_segment[2:].strip()
-    elif text_segment.startswith("3F"):
+        processed_text = processed_text[2:].strip()
+    elif processed_text.startswith("3F"):
         current_voice_short = voice3F_short
-        processed_text = text_segment[2:].strip()
-    elif text_segment.startswith("4F"):
+        processed_text = processed_text[2:].strip()
+    elif processed_text.startswith("4F"):
         current_voice_short = voice4F_short
-        processed_text = text_segment[2:].strip()
-    elif text_segment.startswith("1M"):
+        processed_text = processed_text[2:].strip()
+    elif processed_text.startswith("1M"):
         current_voice_short = voice1_short
-        processed_text = text_segment[2:].strip()
-    elif text_segment.startswith("2M"):
+        processed_text = processed_text[2:].strip()
+    elif processed_text.startswith("2M"):
         current_voice_short = voice2_short
-        processed_text = text_segment[2:].strip()
-    elif text_segment.startswith("3M"):
+        processed_text = processed_text[2:].strip()
+    elif processed_text.startswith("3M"):
         current_voice_short = voice3_short
-        processed_text = text_segment[2:].strip()
-    elif text_segment.startswith("4M"):
+        processed_text = processed_text[2:].strip()
+    elif processed_text.startswith("4M"):
         current_voice_short = voice4_short
-        processed_text = text_segment[2:].strip()
-    elif text_segment.startswith("1O"):  # Old man voice
+        processed_text = processed_text[2:].strip()
+    elif processed_text.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
+        processed_text = processed_text[2:].strip()
+    elif processed_text.startswith("1C"):  #Child voice
         current_voice_short = voice6_short
-        processed_text = text_segment[2:].strip()
+        processed_text = processed_text[2:].strip()
+
+    if processed_text:
+        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
+    return None
+
+async def process_transcript_line(line, default_voice, rate, pitch):
+    """Processes a single transcript line with timestamp and quoted text segments."""
+    match = re.match(r'(\d+):(\d+)(?:\.(\d+))?\s+(.*)', line)
+    if match:
+        minutes, seconds, milliseconds_str, text_parts = match.groups()
+        start_time_ms = int(minutes) * 60000 + int(seconds) * 1000 + (int(milliseconds_str) * 10 if milliseconds_str else 0)
+        audio_segments = []
+        split_parts = re.split(r'(")', text_parts)  # Split by quote marks, keeping the quotes
+
+        process_next = False
+        for part in split_parts:
+            if part == '"':
+                process_next = not process_next
+                continue
+            if process_next and part.strip():
+                audio_path = await generate_audio_with_voice_prefix(part, default_voice, rate, pitch)
+                if audio_path:
+                    audio_segments.append(audio_path)
+            elif not process_next and part.strip():
+                audio_path = await generate_audio_with_voice_prefix(part, default_voice, rate, pitch) # Process unquoted text with default voice
+                if audio_path:
+                    audio_segments.append(audio_path)
 
-    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
+        return start_time_ms, audio_segments
+    return None, None
 
 async def transcript_to_speech(transcript_text, voice, rate, pitch):
     if not transcript_text.strip():
@@ -116,67 +144,42 @@ async def transcript_to_speech(transcript_text, voice, rate, pitch):
     if not voice:
         return None, gr.Warning("Please select a voice.")
 
-    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."
+    lines = transcript_text.strip().split('\n')
+    timed_audio_segments = []
+    max_end_time_ms = 0
+
+    for line in lines:
+        start_time, audio_paths = await process_transcript_line(line, voice, rate, pitch)
+        if start_time is not None and audio_paths:
+            combined_line_audio = AudioSegment.empty()
+            for path in audio_paths:
+                try:
+                    audio = AudioSegment.from_mp3(path)
+                    combined_line_audio += audio
+                    os.remove(path)
+                except FileNotFoundError:
+                    print(f"Warning: Audio file not found: {path}")
+
+            if combined_line_audio:
+                timed_audio_segments.append({'start': start_time, 'audio': combined_line_audio})
+                max_end_time_ms = max(max_end_time_ms, start_time + len(combined_line_audio))
+        elif audio_paths:
+            for path in audio_paths:
+                try:
+                    os.remove(path)
+                except FileNotFoundError:
+                    pass # Clean up even if no timestamp
+
+    if not timed_audio_segments:
+        return None, "No processable audio segments found."
+
+    final_audio = AudioSegment.silent(duration=max_end_time_ms, frame_rate=24000)
+    for segment in timed_audio_segments:
+        final_audio = final_audio.overlay(segment['audio'], position=segment['start'])
+
+    combined_audio_path = tempfile.mktemp(suffix=".mp3")
+    final_audio.export(combined_audio_path, format="mp3")
+    return combined_audio_path, None
 
 @spaces.GPU
 def tts_interface(transcript, voice, rate, pitch):
@@ -187,23 +190,19 @@ async def create_demo():
     voices = await get_voices()
     default_voice = "en-US-AndrewMultilingualNeural - en-US (Male)"
     description = """
-    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.
+    Process timestamped text with voice changes within quotes.
+    Format: `minutes:seconds[.milliseconds] "VoicePrefix Text" more text "AnotherVoicePrefix More Text"`
     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."
-    "and then the default voice continues"
+    0:00 "This is the default voice." more default. "1F Now a female voice." and back to default.
+    0:05 "1C Yes," said the child, "it is fun!"
     ```
     """
     demo = gr.Interface(
         fn=tts_interface,
         inputs=[
-            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.Textbox(label="Timestamped Text with Voice Changes", lines=10, placeholder='0:00 "Text" more text "1F Different Voice"'),
+            gr.Dropdown(choices=[""] + list(voices.keys()), label="Select Default 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)
         ],
@@ -211,7 +210,7 @@ async def create_demo():
             gr.Audio(label="Generated Audio", type="filepath"),
             gr.Markdown(label="Warning", visible=False)
         ],
-        title="Combined TTS: Timestamps and Voice Switching",
+        title="TTS with Timestamp and In-Quote Voice Switching",
         description=description,
         analytics_enabled=False,
         allow_flagging=False