Update app.py

app.py

@@ -33,6 +33,8 @@ import traceback
 from TTS.api import TTS
 import torch
 from TTS.tts.configs.xtts_config import XttsConfig
+from pydub import AudioSegment
+from pyannote.audio import Pipeline
 
 # Accept license terms for Coqui XTTS
 os.environ["COQUI_TOS_AGREED"] = "1"
@@ -128,12 +130,47 @@ logging.basicConfig(level=logging.DEBUG, format="%(asctime)s - %(levelname)s - %
 logger = logging.getLogger(__name__)
 logger.info(f"MoviePy Version: {moviepy.__version__}")
 
+def segment_background_audio(audio_path, output_path="background_segments.wav"):
+
+    # Step 2: Initialize pyannote voice activity detection pipeline (you need Hugging Face token)
+
+    pipeline = Pipeline.from_pretrained(
+        "pyannote/voice-activity-detection",
+        use_auth_token=HUGGINGFACE_TOKEN
+    )
+
+    # Step 3: Run VAD to get speech segments
+    vad_result = pipeline(audio_path)
+    print(f"Detected speech segments: {vad_result}")
+
+    # Step 4: Load full audio and subtract speech segments
+    full_audio = AudioSegment.from_wav(audio_path)
+    background_audio = AudioSegment.silent(duration=len(full_audio))
+
+    for segment in vad_result.itersegments():
+        start_ms = int(segment.start * 1000)
+        end_ms = int(segment.end * 1000)
+        # Remove speech by muting that portion
+        background_audio = background_audio.overlay(AudioSegment.silent(duration=end_ms - start_ms), position=start_ms)
+
+    # Step 5: Subtract background_audio from full_audio
+    result_audio = full_audio.overlay(background_audio)
+
+    # Step 6: Export non-speech segments
+    result_audio.export(output_path, format="wav")
+    print(f"Saved non-speech (background) audio to: {output_path}")
+
+    return output_path
+
 def transcribe_video_with_speakers(video_path):
     # Extract audio from video
     video = VideoFileClip(video_path)
     audio_path = "audio.wav"
     video.audio.write_audiofile(audio_path)
     logger.info(f"Audio extracted from video: {audio_path}")
+
+    segment_background_audio(audio_path)
+    print(f"Saved non-speech (background) audio to local")
     
     # Set up device
     device = "cuda" if torch.cuda.is_available() else "cpu"