app.py

import os
import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
from huggingface_hub import HfApi
import string
import os
from moviepy.editor import VideoFileClip, concatenate_videoclips, ImageClip

huggingface_token = os.getenv('NJOGERERA_TOKEN')
if not huggingface_token:
    raise ValueError("Hugging Face token is not set in the environment variables.")

api = HfApi()
try:
    user_info = api.whoami(token=huggingface_token)
    print(f"Logged in as: {user_info['name']}")
except Exception as e:
    raise ValueError("Failed to authenticate with the provided Hugging Face token.")

model_path = "vertigo23/njogerera_translation_model_V003"
tokenizer = AutoTokenizer.from_pretrained(model_path, use_auth_token=huggingface_token)
model = AutoModelForSeq2SeqLM.from_pretrained(model_path, use_auth_token=huggingface_token)

translator = pipeline("translation", model=model, tokenizer=tokenizer)

prefix = "translate Luganda to English: "

filler_image_path = "alphabet/break.png"

def clean_and_split(text):
    text = text.lower().translate(str.maketrans('', '', string.punctuation))
    return text.split()

def map_word_to_media(word):
    if os.path.exists(f"KSL/{word}.mp4"):
        return [f"KSL/{word}.mp4"]
    else:
        spelled_word_media = [filler_image_path]
        spelled_word_media += [f"alphabet/{letter}.png" for letter in word if os.path.exists(f"alphabet/{letter}.png")]
        spelled_word_media.append(filler_image_path)
        return spelled_word_media

def stitch_media(media_paths):
    clips = []
    for path in media_paths:
        if path.endswith('.mp4'):
            clips.append(VideoFileClip(path))
        elif path.endswith('.png'):
            image_clip = ImageClip(path).set_duration(0.7)
            clips.append(image_clip)
    if not clips:
        raise ValueError("No media files to stitch.")
    
    final_clip = concatenate_videoclips(clips, method="compose")
    final_clip.fps = 24
    final_clip_path = "KSL/final_translation.mp4"
    final_clip.write_videofile(final_clip_path, codec="libx264", fps=24)
    return final_clip_path

def translate_lg_to_en(text):
    lg_input = prefix + text
    translated_text = translator(lg_input)
    english_translation = translated_text[0]['translation_text']
    words = clean_and_split(english_translation)
    media_paths = []
    for word in words:
        media_paths.extend(map_word_to_media(word))
    ksl_path = stitch_media(media_paths)
    return english_translation, ksl_path

# Gradio interface
gr.Interface(
    fn=translate_lg_to_en,
    inputs=gr.Text(),
    outputs=[gr.Textbox(label="English Translation"), gr.Video(label="KSL Sign Language Animation")],
    title="Njogerera Translation App",
    description="Type in a Luganda sentence and see the translation.",
    article="Above is some sample text to test the results of the model. Click to see the results.",
    examples=[
        ["Ebikolwa ebitali bya buntu tebikkirizibwa mu kitundu."],
        ["Olugudo olugenda e Masaka lugadwawo."],
        ["Abalwadde ba Malaria mu dwaliro lye Nsambya bafunye obujanjabi."],
    ],
    allow_flagging="never"
).launch()