app.py

import os
import gradio as gr
from uuid import uuid4
from pipelines.pipeline import InferencePipeline

TITLE = """
    <div style="text-align: center; max-width: 650px; margin: 0 auto;">
        <div
        style="
            display: inline-flex;
            align-items: center;
            gap: 0.8rem;
            font-size: 1.75rem;
        "
        >
        <h1 style="font-weight: 900; margin-bottom: 7px;">
            Auto-AVSR: Audio-Visual Speech Recognition
        </h1>
        </div>
        <p style="margin-bottom: 10px; font-size: 94%">
        Want to recognize content in a noisy environment?<br>Our Auto-AVSR models are here to transcribe your answers from audio or visual information!
        </p>
    </div>
"""

ARTICLE = """
<div style="text-align: center; max-width: 650px; margin: 0 auto;">
    <p>
    Want to look into models? You can find our [<a href="https://github.com/mpc001/auto_avsr">training code</a>] and [<a href="https://arxiv.org/abs/2303.14307">paper</a>].
    </p>
    <p>
    The inference is performed on the CPU. You can also run on <a href="https://colab.research.google.com/drive/1jfb6e4xxhXHbmQf-nncdLno1u0b4j614?usp=sharing">Colab GPU</a>
    </p>
    <p>
    We share this demo only for non-commercial purposes.
    </p>
</div>
"""

CSS = """
    #col-container {margin-left: auto; margin-right: auto;}
    a {text-decoration-line: underline; font-weight: 600;}
    .animate-spin {
        animation: spin 1s linear infinite;
    }
    @keyframes spin {
        from { transform: rotate(0deg); }
        to { transform: rotate(360deg); }
    }
    #share-btn-container {
        display: flex; padding-left: 0.5rem !important; padding-right: 0.5rem !important; background-color: #000000; justify-content: center; align-items: center; border-radius: 9999px !important; width: 13rem;
    }
    #share-btn {
        all: initial; color: #ffffff;font-weight: 600; cursor:pointer; font-family: 'IBM Plex Sans', sans-serif; margin-left: 0.5rem !important; padding-top: 0.25rem !important; padding-bottom: 0.25rem !important;
    }
    #share-btn * {
        all: unset;
    }
    #share-btn-container div:nth-child(-n+2){
        width: auto !important;
        min-height: 0px !important;
    }
    #share-btn-container .wrap {
        display: none !important;
    }
"""

FFMPEG_COMMAND = "-loglevel error -y -r 25 -pix_fmt yuv420p -f mp4"

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")
}

def fn(pipeline_type, filename):
    directory = "./tmp"
    if not os.path.exists(directory):
        os.makedirs(directory)
    dst_filename = os.path.join(directory, str(uuid4())[:8]+".mp4")
    command_string = f"ffmpeg -i {filename} {FFMPEG_COMMAND} {dst_filename}"
    os.system(command_string)
    selected_pipeline_instance = pipelines[pipeline_type]
    landmarks = selected_pipeline_instance.process_landmarks(dst_filename, landmarks_filename=None)
    data = selected_pipeline_instance.dataloader.load_data(dst_filename, landmarks)
    transcript = selected_pipeline_instance.model.infer(data)
    return transcript

demo = gr.Blocks(css=CSS)

with demo:
    gr.HTML(TITLE)
    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)
    gr.HTML(ARTICLE)
demo.launch()