utils/audio_palette.py

import typing
from datetime import datetime

import PIL
from PIL import Image
from moviepy.editor import *

from lib import *

datetime_format = "%d/%m/%Y %H:%M:%S"
def now():
    return datetime.now().strftime(datetime_format)

class AudioPalette:
    def __init__(self, pace_model_weights_path, resnet50_tf_model_weights_path, height, width, channels):
        self.pace_model = PaceModel(height, width, channels, resnet50_tf_model_weights_path, pace_model_weights_path)
        self.image_captioning = ImageCaptioning()
        self.audio_generation = AudioGeneration()
        self.sentiment_analyser = SentimentAnalyser()
        self.pace_map = {
            "Fast": "high",
            "Medium": "medium",
            "Slow": "low"
        }
    
    def prompt_construction(self, caption: str, pace: str, sentiment: typing.Union[str, None], instrument: typing.Union[str, None], first: bool = True):
        instrument = instrument if instrument is not None else ""

        if first:
            prompt = f"A {instrument} soundtrack for {caption} with {self.pace_map[pace]} beats per minute. High Quality."
        else:
            prompt = f"A {instrument} soundtrack for {caption} with {self.pace_map[pace]} beats per minute. High Quality. Transitions smoothely from the previous audio while sounding different."
        
        # if sentiment:
        #     prompt += f" As a {sentiment} music."
        
        return prompt
    
    def generate_single(self, input_image: PIL.Image.Image, instrument: typing.Union[str, None], ngrok_endpoint: str):
        pace = self.pace_model.predict(input_image)
        print(f"[{now()}]", pace)
        print(f"[{now()}] Pace Prediction Done")

        generated_text = self.image_captioning.query(input_image)[0].get("generated_text")
        print(f"[{now()}]", generated_text)
        print(f"[{now()}] Captioning Done")

        sentiment = self.sentiment_analyser.sentiment(generated_text)
        print(f"[{now()}] Sentiment Analysis Done")

        prompt = self.prompt_construction(generated_text, pace, sentiment, instrument)
        print(f"[{now()}] Generated Prompt:", prompt)

        audio_file = self.audio_generation.generate(prompt, ngrok_endpoint)
        print(f"[{now()}]", audio_file)
        print(f"[{now()}] Audio Generation Done")

        outputs = [prompt, pace, generated_text, audio_file]
        return outputs
    
    def stitch_images(self, file_paths: typing.List[str], audio_paths: typing.List[str]):
        clips = [ImageClip(m).set_duration(5) for m in file_paths]
        audio_clips = [AudioFileClip(a) for a in audio_paths]
        concat_audio = concatenate_audioclips(audio_clips)
        new_audio = CompositeAudioClip([concat_audio])

        concat_clip = concatenate_videoclips(clips, method="compose")
        concat_clip.audio = new_audio
        
        file_name = "generated_video.mp4"
        concat_clip.write_videofile(file_name, fps=24)
        return file_name
    
    def generate_multiple(self, file_paths: typing.List[str], instrument: typing.Union[str, None], ngrok_endpoint: str):
        images = [Image.open(image_path) for image_path in file_paths]
        pace = []
        generated_text = []
        sentiments = []
        prompts = []

        # Extracting the pace for all the images
        for image in images:
            pace_prediction = self.pace_model.predict(image)
            pace.append(pace_prediction)
        print(f"[{now()}]", pace)
        print(f"[{now()}] Pace Prediction Done")
        
        # Generating the caption for all the images
        for image in images:
            caption = self.image_captioning.query(image)[0].get("generated_text")
            generated_text.append(caption)
        print(f"[{now()}]", generated_text)
        print(f"[{now()}] Captioning Done")

        # Extracting the sentiments from the generated captions
        for text in generated_text:
            sentiment = self.sentiment_analyser.sentiment(text)
            sentiments.append(sentiment)
        print(f"[{now()}] Sentiment Analysis Done:", sentiments)

        first = True
        for generated_caption, senti, pace_pred in zip(generated_text, sentiments, pace):
            prompts.append(self.prompt_construction(generated_caption, pace_pred, senti, instrument, first))
            first = False
        print(f"[{now()}] Generated Prompts:", prompts)

        audio_file = self.audio_generation.generate(prompts, ngrok_endpoint)
        print(f"[{now()}]", audio_file)
        print(f"[{now()}] Audio Generation Done")

        video_file = self.stitch_images(file_paths, [audio_file])
        return video_file