verify voice offloaded to Modal

app.py

@@ -1,11 +1,12 @@
 import gradio as gr
-from detector.face import verify_faces, analyze_face
 from detector.voice import verify_voices
 from detector.video import verify_faces_in_video
 from reports.pdf_report import generate_pdf_report
 from utils.youtube_utils import download_youtube_video
 import modal
 verify_faces_remote = modal.Function.lookup("deepface-agent", "verify_faces_remote")
+verify_voices_remote = modal.Function.lookup("deepface-agent", "verify_voices_remote")
+
 
 
 # Holds latest results
@@ -13,7 +14,6 @@ last_face_result = None
 last_voice_result = None
 last_video_results = None
 
-# @app.local_entrypoint()
 def compare_faces(img1_path: str, img2_path: str) -> str:
     """Use this tool to compare to faces for a match
     Args:
@@ -51,20 +51,28 @@ def compare_voices(audio1: str, audio2: str) -> str:
         audio2: The path to the second audio file
     """
     global last_voice_result
-    result = verify_voices(audio1, audio2)
-    result_text = ""
+    
+    try:
+        with open(audio1, "rb") as a1, open(audio2, "rb") as a2:
+            audio1_bytes = a1.read()
+            audio2_bytes = a2.read()
+    
+        result = verify_voices_remote.remote(audio1_bytes, audio2_bytes)
+        result_text = ""
 
-    if "error" in result:
-        return f"❌ Error: {result['error']}"
+        if "error" in result:
+            return f"❌ Error: {result['error']}"
 
-    if result["match"]:
-        result_text = f"✅ Same speaker detected. Similarity: {result['similarity']} (Threshold: {result['threshold']})"
-        last_voice_result = result_text
-        return result_text
-    else:
-        result_text = f"❌ Different speakers. Similarity: {result['similarity']} (Threshold: {result['threshold']})"
-        last_voice_result = result_text
-        return result_text
+        if result["match"]:
+            result_text = f"✅ Same speaker detected. Similarity: {result['similarity']} (Threshold: {result['threshold']})"
+            last_voice_result = result_text
+            return result_text
+        else:
+            result_text = f"❌ Different speakers. Similarity: {result['similarity']} (Threshold: {result['threshold']})"
+            last_voice_result = result_text
+            return result_text
+    except Exception as e:
+        return f"❌ Error reading audio files: {str(e)}"
 
 
 def scan_video(video_file: str, ref_img: str, youtube_url="") -> str:

modal_app/modal_app.py

@@ -0,0 +1,79 @@
+import modal
+
+app = modal.App("deepface-agent")
+
+# Container Image
+image = (
+    modal.Image.debian_slim()
+    .apt_install("libgl1", "libglib2.0-0")
+    .pip_install(
+        "deepface",
+        "opencv-python",
+        "numpy",
+        "Pillow",
+        "tensorflow==2.19.0",
+        "tf-keras>=2.19.0",
+        "librosa",
+        "scipy",
+        "speechbrain",
+        "torchaudio",
+    )
+)
+
+# ✅ This block runs *inside* the Modal container only
+# To void repeatedly loading the model on each function call
+with image.imports():
+    from speechbrain.pretrained import SpeakerRecognition
+
+    # Model for voice recognition
+    verification = SpeakerRecognition.from_hparams(
+        source="speechbrain/spkrec-ecapa-voxceleb",
+        savedir="pretrained_models/spkrec-ecapa-voxceleb",
+    )
+
+
+@app.function(image=image, gpu="any")
+def verify_faces_remote(img1_bytes, img2_bytes):
+    """
+    Accepts images bytes and compare them for a match.
+    """
+    from deepface import DeepFace
+    from PIL import Image
+    from io import BytesIO
+    import numpy as np
+
+    img1 = np.array(Image.open(BytesIO(img1_bytes)))
+    img2 = np.array(Image.open(BytesIO(img2_bytes)))
+
+    result = DeepFace.verify(img1, img2)
+    return result
+
+
+@app.function(image=image, gpu="any")
+def verify_voices_remote(audio1_bytes, audio2_bytes):
+    """
+    Accepts audio bytes and compare them for a match.
+    """
+
+    import tempfile
+    import pathlib
+
+    with (
+        tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f1,
+        tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f2,
+    ):
+        f1.write(audio1_bytes)
+        f2.write(audio2_bytes)
+        audio1_path = f1.name
+        audio2_path = f2.name
+
+    try:
+        score, prediction = verification.verify_files(audio1_path, audio2_path)
+
+        return {"match": prediction, "similarity": float(score), "threshold": 0.75}
+    except Exception as e:
+        return {"error": str(e)}
+
+    finally:
+        pathlib.Path(audio1_path).unlink()
+        pathlib.Path(audio2_path).unlink()