Update app.py

app.py

@@ -102,29 +102,20 @@ async def generate_audio_with_voice_prefix(text_segment, default_voice, rate, pi
             with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
                 audio_path = tmp_file.name
                 await communicate.save(audio_path)
-            if target_duration_ms is not None and os.path.exists(audio_path):
+            if target_duration_ms is not None and os.path.exists(audio_path) and target_duration_ms > 0:
                 audio = AudioSegment.from_mp3(audio_path)
                 audio_duration_ms = len(audio)
                 #print(f"Generated audio duration: {audio_duration_ms}ms, Target duration: {target_duration_ms}ms") # Debug
 
-                if audio_duration_ms > target_duration_ms and target_duration_ms > 0:
+                if audio_duration_ms > target_duration_ms:
                     speed_factor = (audio_duration_ms / target_duration_ms) * speed_adjustment_factor
                     #print(f"Speed factor (after user adjustment): {speed_factor}") # Debug
                     if speed_factor > 0:
                         if speed_factor < 1.0:
                             speed_factor = 1.0
-                        #y, sr = librosa.load(audio_path, sr=None)
-                        
-                        # Load audio file
                         audio = AudioSegment.from_file(audio_path)
-                        # Apply time-stretching
                         audio_stretched = audio.speedup(playback_speed=speed_factor)
-                        # Save the stretched audio
                         audio_stretched.export(audio_path, format="mp3")
-
-                        #y_stretched = librosa.effects.time_stretch(y, rate=speed_factor)
-                        #sf.write(audio_path, y_stretched, sr) 
-                        
                 else:
                     print("Generated audio is not longer than target duration, no speed adjustment.") # Debug
             return audio_path
@@ -133,24 +124,21 @@ async def generate_audio_with_voice_prefix(text_segment, default_voice, rate, pi
             return None
     return None
 
-async def process_transcript_line(line, default_voice, rate, pitch, speed_adjustment_factor):
-    """Processes a single transcript line with HH:MM:SS,milliseconds - HH:MM:SS,milliseconds timestamp."""
-    match = re.match(r'(\d{2}):(\d{2}):(\d{2}),(\d{3})\s+-\s+(\d{2}):(\d{2}):(\d{2}),(\d{3})\s+(.*)', line)
+async def process_transcript_line(line, next_line_start_time, default_voice, rate, pitch, speed_adjustment_factor):
+    """Processes a single transcript line with HH:MM:SS,milliseconds timestamp."""
+    match = re.match(r'(\d{2}):(\d{2}):(\d{2}),(\d{3})\s+(.*)', line)
     if match:
-        start_h, start_m, start_s, start_ms, end_h, end_m, end_s, end_ms, text_parts = match.groups()
+        start_h, start_m, start_s, start_ms, text_parts = match.groups()
         start_time_ms = (
             int(start_h) * 3600000 +
             int(start_m) * 60000 +
             int(start_s) * 1000 +
             int(start_ms)
         )
-        end_time_ms = (
-            int(end_h) * 3600000 +
-            int(end_m) * 60000 +
-            int(end_s) * 1000 +
-            int(end_ms)
-        )
-        duration_ms = end_time_ms - start_time_ms
+        duration_ms = None
+        if next_line_start_time is not None:
+            duration_ms = next_line_start_time - start_time_ms
+
         audio_segments = []
         split_parts = re.split(r'[“”"]', text_parts)
         process_next = False
@@ -177,12 +165,22 @@ async def transcript_to_speech(transcript_text, voice, rate, pitch, speed_adjust
     lines = transcript_text.strip().split('\n')
     timed_audio_segments = []
     max_end_time_ms = 0
-    for line in lines:
-        start_time, audio_paths, duration = await process_transcript_line(line, voice, rate, pitch, speed_adjustment_factor)
+    for i, line in enumerate(lines):
+        next_line_start_time = None
+        if i < len(lines) - 1:
+            next_line_match = re.match(r'(\d{2}):(\d{2}):(\d{2}),(\d{3})\s+.*', lines[i+1])
+            if next_line_match:
+                nh, nm, ns, nms = next_line_match.groups()
+                next_line_start_time = (
+                    int(nh) * 3600000 +
+                    int(nm) * 60000 +
+                    int(ns) * 1000 +
+                    int(nms)
+                )
+
+        start_time, audio_paths, duration = await process_transcript_line(line, next_line_start_time, voice, rate, pitch, speed_adjustment_factor)
         if start_time is not None and audio_paths:
             combined_line_audio = AudioSegment.empty()
-            current_time_ms = start_time
-            segment_duration = duration / len(audio_paths) if audio_paths else 0
             for path in audio_paths:
                 if path:  # Only process if audio_path is not None (meaning TTS was successful)
                     try:
@@ -201,11 +199,14 @@ async def transcript_to_speech(transcript_text, voice, rate, pitch, speed_adjust
                         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
@@ -219,14 +220,16 @@ async def create_demo():
     voices = await get_voices()
     default_voice = "en-US-AndrewMultilingualNeural - en-US (Male)"
     description = """
-    Process timestamped text (HH:MM:SS,milliseconds - HH:MM:SS,milliseconds) with voice changes within quotes.
-    The duration specified in the timestamp will be used to adjust the speech rate so the generated audio fits within that time.
+    Process timestamped text (HH:MM:SS,milliseconds) with voice changes within quotes.
+    The duration for each segment is determined by the timestamp of the following line.
+    The speed of the generated audio will be adjusted to fit within this duration.
+    If there is no subsequent timestamp, the speed adjustment will be skipped.
     You can control the intensity of the speed adjustment using the "Speed Adjustment Factor" slider.
-    Format: `HH:MM:SS,milliseconds - HH:MM:SS,milliseconds "VoicePrefix Text" more text "AnotherVoicePrefix More Text"`
+    Format: `HH:MM:SS,milliseconds "VoicePrefix Text" more text "AnotherVoicePrefix More Text"`
     Example:
     ```
-    00:00:00,000 - 00:00:05,000 "This is the default voice." more default. "1F Now a female voice." and back to default.
-    00:00:05,500 - 00:00:10,250 "1C Yes," said the child, "it is fun!"
+    00:00:00,000 "This is the default voice." more default. "1F Now a female voice." and back to default.
+    00:00:05,500 "1C Yes," said the child, "it is fun!"
     ```
     ***************************************************************************************************
     1M = en-AU-WilliamNeural - en-AU (Male)
@@ -248,7 +251,7 @@ async def create_demo():
     demo = gr.Interface(
         fn=tts_interface,
         inputs=[
-            gr.Textbox(label="Timestamped Text with Voice Changes and Duration", lines=10, placeholder='00:00:00,000 - 00:00:05,000 "Text" more text "1F Different Voice"'),
+            gr.Textbox(label="Timestamped Text with Voice Changes and Duration", lines=10, placeholder='00:00:00,000 "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),
@@ -258,7 +261,7 @@ async def create_demo():
             gr.Audio(label="Generated Audio", type="filepath"),
             gr.Markdown(label="Warning", visible=False)
         ],
-        title="TTS with Duration-Aware Speed Adjustment and In-Quote Voice Switching",
+        title="TTS with Dynamic Duration and In-Quote Voice Switching",
         description=description,
         analytics_enabled=False,
         allow_flagging=False