Create app.py

app.py

@@ -0,0 +1,127 @@
+import gradio as gr
+import gradio as gr
+import torch
+import soundfile as sf
+import io
+import librosa
+import numpy as np
+from pytube import YouTube
+import os
+import random
+
+FS=16000
+MAX_SIZE = FS * 30
+MODEL = torch.jit.load('model.pt')
+
+
+def reformat_freq(sr, y):
+    if len(y.shape)==1 or y.shape[1]==1:
+        pass
+        #print("monocanal")
+    else:
+        # Avg two channels
+        y=y.mean(axis=1)
+
+    y = y.astype(np.float32)
+    if sr not in (
+        FS,
+    ):
+        y = librosa.resample(y, orig_sr=sr, target_sr=FS)
+
+    return sr, y
+
+
+def preprocess_audio(audio):
+    _, y = reformat_freq(*audio)
+    y = y[:MAX_SIZE]
+    y=torch.as_tensor(y,dtype=torch.float32)
+    y=torch.unsqueeze(y,0)
+
+    return y
+
+def postprocess_output(score):
+    out=score.item()
+    out = round(100*out,2)
+    return "{:.2f}%".format(out)
+
+def process_youtube_address(youtube_address):
+    print("Downloading youtube audio from video...")
+    
+    try:
+        selected_video = YouTube(youtube_address)
+        audio=selected_video.streams.filter(only_audio=True, file_extension='mp4').first()
+        nrand=round(random.random()*1000)
+        audioname="audio-"+str(nrand)+".mp4a"
+        audiowav="audio-"+str(nrand)+".wav"
+        audiomp4a=audio.download('tmp',audioname)
+        os.system("ffmpeg -i " + audiomp4a + " -ac 1 -ar {} ".format(FS) + audiowav + "; rm tmp/" + audioname )
+    except Exception as inst:
+        print("Exception: {}".format(inst))
+        print("ERROR while downloading audio from " + youtube_address)
+        audiowav=None
+    return audiowav
+
+
+
+
+def process_micro(micro):
+    x=preprocess_audio(micro)
+    output = MODEL(x)
+    print(output)
+    result = postprocess_output(output)
+
+    return result
+
+def process_file(file):
+    x,fs = librosa.load(file, sr=FS)
+    x=preprocess_audio((fs,x))
+    print("Running model")
+    output = MODEL(x)
+    print(output)
+    result = postprocess_output(output)
+
+    return result
+
+def process_youtube(youtube_address):
+    audiofile=process_youtube_address(youtube_address)
+
+    if audiofile is not None:
+        result = process_file(audiofile)
+        return result
+    else: 
+        return "Could not get audio from {}".format(youtube_address)
+
+
+
+
+
+
+with gr.Blocks(title="Audio Fake Detector") as demo:
+    gr.Markdown("# Welcome to Loccus.ai synthetic voice detection demo!")
+    
+
+    
+    with gr.Row():
+        with gr.Column():
+            m = gr.Audio(source="microphone", type="numpy",label="Micro")
+            f = gr.Audio(source="upload", type="filepath", label="Audio file")
+            y = gr.Textbox(label="Enter YouTube address here")
+
+
+        with gr.Column():
+            with gr.Row(equal_height=True):
+            
+                text = gr.Textbox(label="Probability of Real Voice")
+        
+        #file= gr.Audio(source="upload", type="filepath", optional=True)
+    button_clear = gr.ClearButton([m,f,y,text])
+    m.stop_recording(process_micro, inputs=[m], outputs=text)
+    f.upload(process_file,inputs=[f], outputs=text)
+    y.submit(process_youtube, inputs=[y], outputs=text)
+
+
+    #btn = gr.Button("Run")
+    #btn.click(fn=update, inputs=inp, outputs=out)
+
+demo.launch()
+