hD13's picture
Update app.py
956bd26
import gradio as gr
from transformers import pipeline
import cv2
from PIL import Image
import io
import scipy
import torch
import time
def video_to_descriptions(video, target_language="en"):
start_time = time.time()
print("START TIME = ", start_time)
ImgToText = pipeline("image-to-text", model="Salesforce/blip-image-captioning-large")
Summarize = pipeline("summarization", model="tuner007/pegasus_summarizer")
translator = pipeline("translation", model=f"Helsinki-NLP/opus-mt-en-{target_language}")
audio = pipeline("text-to-speech", model="suno/bark-small")
voice_preset = f"v2/{target_language}_speaker_1"
cap = cv2.VideoCapture(video)
fps = int(cap.get(cv2.CAP_PROP_FPS))
descriptions = []
frame_count = 0
while True:
ret, frame = cap.read()
if not ret:
break
if frame_count % (fps * 2) == 0:
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
pil_img = Image.fromarray(frame_rgb)
outputs = ImgToText(pil_img)
description = outputs[0]['generated_text']
descriptions.append(description)
print(str(frame_count) + " : " + outputs[0]['generated_text'])
frame_count += 1
cap.release()
concatenated_description = " ".join(descriptions)
summarized_description = Summarize(concatenated_description, max_length=31)[0]["summary_text"]
print("SUMMARIZATION : " + summarized_description)
translated_text = translator(summarized_description)[0]["translation_text"]
print("TRANSLATION : " + translated_text)
audio_file = audio(translated_text)
output_path = "./bark_out.wav"
scipy.io.wavfile.write(output_path, data=audio_file["audio"][0], rate=audio_file["sampling_rate"])
stop_time = time.time()
print("EXECUTION TIME = ", stop_time - start_time)
return output_path
language_dropdown = gr.Dropdown(
["en", "fr", "de", "es"], label="[MANDATORY] Language", info="The Voice's Language"
)
iface = gr.Interface(
fn=video_to_descriptions,
inputs=[gr.Video(label="Video to Upload", info="The Video"), language_dropdown],
outputs="audio",
live=False
)
if __name__ == "__main__":
iface.launch()