Update app.py

app.py

@@ -1,116 +1,59 @@
 import gradio as gr
-# import os, subprocess, torchaudio
-# import torch
-from PIL import Image
-from gtts import gTTS
-import tempfile
-from pydub import AudioSegment
-from pydub.generators import Sine
-# from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub
-# from fairseq.models.text_to_speech.hub_interface import TTSHubInterface
-import soundfile
-
-import dlib
-import cv2
-import imageio
-import os
-import gradio as gr
 import os, subprocess, torchaudio
+import torch
 from PIL import Image
-import ffmpeg
-
-
 
 block = gr.Blocks()
 
-
-    
-def one_shot(image,input_text,gender): 
-   if gender == 'Female' or gender == 'female':
-     tts = gTTS(input_text)
-     with tempfile.NamedTemporaryFile(suffix='.mp3', delete=False) as f:
-          tts.write_to_fp(f)
-          f.seek(0)
-          sound = AudioSegment.from_file(f.name, format="mp3")
-          sound.export("/content/audio.wav", format="wav")
-     waveform, sample_rate = torchaudio.load("/content/audio.wav")
-
-     
-     torchaudio.save("/content/audio.wav", waveform, sample_rate, encoding="PCM_S", bits_per_sample=16)
-     image = Image.open(image_in)
-     image = pad_image(image)
-     image.save("/content/image_pre.png")
-     pocketsphinx_run = subprocess.run(['pocketsphinx', '-phone_align', 'yes', 'single', '/content/audio.wav'], check=True, capture_output=True)
-     jq_run = subprocess.run(['jq', '[.w[]|{word: (.t | ascii_upcase | sub("<S>"; "sil") | sub("<SIL>"; "sil") | sub("\\\(2\\\)"; "") | sub("\\\(3\\\)"; "") | sub("\\\(4\\\)"; "") | sub("\\\[SPEECH\\\]"; "SIL") | sub("\\\[NOISE\\\]"; "SIL")), phones: [.w[]|{ph: .t | sub("\\\+SPN\\\+"; "SIL") | sub("\\\+NSN\\\+"; "SIL"), bg: (.b*100)|floor, ed: (.b*100+.d*100)|floor}]}]'], input=pocketsphinx_run.stdout, capture_output=True)
-     with open("test.json", "w") as f:
-         f.write(jq_run.stdout.decode('utf-8').strip())
-     import json
-
-     with open('test.json') as user_file:
-       file_contents = user_file.read()
-      
-     
-    
-     parsed_json = json.loads(file_contents)
-     return parsed_json
-     exit()
-     os.system(f"cd /content/one-shot-talking-face && python3 -B test_script.py --img_path /content/image_pre.png --audio_path /content/audio.wav --phoneme_path /content/test.json --save_dir /content/train")
-     
-       
-     
-       
-   elif gender == 'Male' or gender == 'male':
-      print(gender)
-      models, cfg, task = load_model_ensemble_and_task_from_hf_hub(
-          "Voicemod/fastspeech2-en-male1",
-          arg_overrides={"vocoder": "hifigan", "fp16": False}
-      )
-
-      model = models[0].cuda()
-      TTSHubInterface.update_cfg_with_data_cfg(cfg, task.data_cfg)
-      generator = task.build_generator([model], cfg)
-      # next(model.parameters()).device
-
-      sample = TTSHubInterface.get_model_input(task, input_text)
-      sample["net_input"]["src_tokens"] = sample["net_input"]["src_tokens"].cuda()
-      sample["net_input"]["src_lengths"] = sample["net_input"]["src_lengths"].cuda()
-      sample["speaker"] = sample["speaker"].cuda()
-
-      wav, rate = TTSHubInterface.get_prediction(task, model, generator, sample)
-      # soundfile.write("/content/audio_before.wav", wav, rate)
-      soundfile.write("/content/audio_before.wav", wav.cpu().clone().numpy(), rate)
-      cmd='ffmpeg -i /content/audio_before.wav -filter:a "atempo=0.7" -vn /content/audio.wav'
-      os.system(cmd)
-      one_shot_talking(image,'audio.wav')
-
-       
-
-
-def generate_ocr(method,image,gender):
-    return "Hello"
+def pad_image(image):
+    w, h = image.size
+    if w == h:
+        return image
+    elif w > h:
+        new_image = Image.new(image.mode, (w, w), (0, 0, 0))
+        new_image.paste(image, (0, (w - h) // 2))
+        return new_image
+    else:
+        new_image = Image.new(image.mode, (h, h), (0, 0, 0))
+        new_image.paste(image, ((h - w) // 2, 0))
+        return new_image
+
+def calculate(image_in, audio_in):
+    waveform, sample_rate = torchaudio.load(audio_in)
+    waveform = torch.mean(waveform, dim=0, keepdim=True)
+    torchaudio.save("/content/audio.wav", waveform, sample_rate, encoding="PCM_S", bits_per_sample=16)
+    image = Image.open(image_in)
+    image = pad_image(image)
+    image.save("image.png")
+
+    pocketsphinx_run = subprocess.run(['pocketsphinx', '-phone_align', 'yes', 'single', '/content/audio.wav'], check=True, capture_output=True)
+    jq_run = subprocess.run(['jq', '[.w[]|{word: (.t | ascii_upcase | sub("<S>"; "sil") | sub("<SIL>"; "sil") | sub("\\\(2\\\)"; "") | sub("\\\(3\\\)"; "") | sub("\\\(4\\\)"; "") | sub("\\\[SPEECH\\\]"; "SIL") | sub("\\\[NOISE\\\]"; "SIL")), phones: [.w[]|{ph: .t | sub("\\\+SPN\\\+"; "SIL") | sub("\\\+NSN\\\+"; "SIL"), bg: (.b*100)|floor, ed: (.b*100+.d*100)|floor}]}]'], input=pocketsphinx_run.stdout, capture_output=True)
+    with open("test.json", "w") as f:
+        f.write(jq_run.stdout.decode('utf-8').strip())
+
+    os.system(f"cd /content/one-shot-talking-face && python3 -B test_script.py --img_path /content/image.png --audio_path /content/audio.wav --phoneme_path /content/test.json --save_dir /content/train")
+    return "/content/train/image_audio.mp4"
     
 def run():
   with block:
-  
+    gr.Markdown(
+    """
+    <style> body { text-align: right} </style>
+    map: 📄 [arxiv](https://arxiv.org/abs/2112.02749) &nbsp; ⇨ 👩‍💻 [github](https://github.com/FuxiVirtualHuman/AAAI22-one-shot-talking-face) &nbsp; ⇨ 🦒 [colab](https://github.com/camenduru/one-shot-talking-face-colab) &nbsp; ⇨ 🤗 [huggingface](https://huggingface.co/spaces/camenduru/one-shot-talking-face) &nbsp; | &nbsp; tools: 🌀 [duplicate this space](https://huggingface.co/spaces/camenduru/sandbox?duplicate=true) &nbsp; | 🐢 [tortoise tts](https://huggingface.co/spaces/mdnestor/tortoise) &nbsp; | 📺 [video upscaler](https://huggingface.co/spaces/kadirnar/Anime4k) &nbsp; | 🎨 [text-to-image](https://huggingface.co/models?pipeline_tag=text-to-image&sort=downloads) &nbsp; | 🐣 [twitter](https://twitter.com/camenduru) &nbsp; | ☕ [buy-a-coffee](https://ko-fi.com/camenduru) &nbsp;
+    """)
     with gr.Group():
       with gr.Box():
         with gr.Row().style(equal_height=True):
           image_in = gr.Image(show_label=False, type="filepath")
-          # audio_in = gr.Audio(show_label=False, type='filepath')
-          input_text=gr.Textbox(lines=3, value="Hello How are you?", label="Input Text")
-          gender = gr.Radio(["Female","Male"],value="Female",label="Gender")
-          video_out = gr.Textbox(label="output")
-          # video_out = gr.Video(show_label=False)
+          audio_in = gr.Audio(show_label=False, type='filepath')
+          video_out = gr.Video(show_label=False)
         with gr.Row().style(equal_height=True):
           btn = gr.Button("Generate")          
 
-    btn.click(one_shot, inputs=[image_in, input_text,gender], outputs=[video_out])
-    # block.queue()
+    
+    btn.click(calculate, inputs=[image_in, audio_in], outputs=[video_out])
+    block.queue()
     block.launch(server_name="0.0.0.0", server_port=7860)
 
 if __name__ == "__main__":
-    run()
-
-
-
-
+    run()