app.py

import os
import gradio as gr
from pipelines.pipeline import InferencePipeline

pipelines = {
    "VSR(mediapipe)": InferencePipeline("./configs/LRS3_V_WER19.1.ini", device="cpu", face_track=True, detector="mediapipe"),
    "ASR": InferencePipeline("./configs/LRS3_A_WER1.0.ini", device="cpu", face_track=True, detector="mediapipe"),
    "AVSR(mediapipe)": InferencePipeline("./configs/LRS3_AV_WER0.9.ini", device="cpu", face_track=True, detector="mediapipe")
}
print("Step 0. Model has been loaded.")

def fn(pipeline_type, filename):
    print("Step 0. Video has been uploaded.")
    selected_pipeline_instance = pipelines[pipeline_type]
    print("Step 1. Video has been converted.")
    landmarks = selected_pipeline_instance.process_landmarks(filename, landmarks_filename=None)
    print("Step 2. Landmarks have been detected.")
    data = selected_pipeline_instance.dataloader.load_data(filename, landmarks)
    print("Step 3. Data has been preprocessed.")
    transcript = selected_pipeline_instance.model.infer(data)
    print("Step 4. Inference has been done.")
    print(f"transcript: {transcript}")
    return transcript

demo = gr.Blocks()

with demo:

    gr.HTML(
        """
        <div style="text-align: center; max-width: 1200px; margin: 20px auto;">
        <h1 style="font-weight: 900; font-size: 3rem; margin: 0rem">
            Auto-AVSR
        </h1>        
        <h3 style="font-weight: 450; font-size: 1rem; margin: 0rem"> 
        [<a href="https://arxiv.org/abs/2303.14307" style="color:blue;">arXiv</a>] 
        [<a href="https://github.com/mpc001/auto_avsr" style="color:blue;">Code</a>]
        </h3> 
        <h2 style="text-align: left; font-weight: 450; font-size: 1rem; margin-top: 0.5rem; margin-bottom: 0.5rem">
        🔥 <b>Notes</b>: We share this demo only for non-commercial purposes.
        </h2>
        
        </div>
        """)

    
    dropdown_list = gr.inputs.Dropdown(["ASR", "VSR(mediapipe)", "AVSR(mediapipe)"], label="model")
    video_file = gr.Video(label="INPUT VIDEO", include_audio=True)
    text = gr.Textbox(label="PREDICTION")
    btn = gr.Button("Submit").style(full_width=True)

    btn.click(fn, inputs=[dropdown_list, video_file], outputs=text)

demo.launch()