Support for longer videos

app.py

@@ -1,5 +1,6 @@
 import gradio as gr
 from typing import Any
+import math
 
 import torch
 from transformers import pipeline
@@ -14,6 +15,9 @@ from textprocessor import TextProcessor
 from videocreator import VideoCreator
 from share_btn import community_icon_html, loading_icon_html, share_js
 
+MAX_NUM_WORDS = 20000
+MAX_CHUNK_LENGTH = 600
+
 spanish_transcribe_model = "juancopi81/whisper-medium-es"
 languages = {"Spanish": "es", "English": "en"}
 
@@ -82,8 +86,28 @@ def datapipeline(url: str,
         return "Language not supported"
     transcribed_text = audio_transcriber.transcribe(audio_path_file)
     print("Audio transcription ready!")
-    json_scenes = text_processor.get_json_scenes(transcribed_text,
-                                                 summary_language)
+    # Get total number of words in text
+    num_words_transcription = len(transcribed_text.split())
+
+    if num_words_transcription > MAX_NUM_WORDS:
+        print("to add return here")
+
+    if num_words_transcription > MAX_CHUNK_LENGTH:
+        num_chunks = math.ceil(num_words_transcription / MAX_CHUNK_LENGTH)
+        num_words_per_chunk = num_words_transcription // num_chunks
+        chunks = utils.splitter(num_words_per_chunk, transcribed_text)
+        json_scenes = {}
+        for chunk in chunks:
+            if len(chunk.split()) > 50:
+                max_key = max(json_scenes.keys(), default=0)
+                chunk_scenes = text_processor.get_json_scenes(chunk,
+                                                              summary_language)
+                chunk_scenes = {k+max_key: v for k, v in chunk_scenes.items()}
+                json_scenes.update(chunk_scenes)
+    else:
+        json_scenes = text_processor.get_json_scenes(transcribed_text,
+                                                     summary_language)
+
     print("Scenes ready")
     video = video_creator.create_video(json_scenes, video_styles)
     print("Video at", video)

textprocessor.py

@@ -9,10 +9,10 @@ You are a creator of illustrated books building a series of scenes for your book
 Your boss asked you to write a summary and illustrations of this text:
 $TRANSCRIPTION
 You have to write the summary using a maximum of 7 scenes in a JSON object following these instructions:
-[{Scene": int, "Summary": $SUMMARY_LANGUAGE str, "Illustration": English str}, ...] where:
+[{"Scene": int, "Summary": "$SUMMARY_LANGUAGE str", "Illustration": "English str"}, ...] where:
 "Scene": The number of the scene.
-"Summary": $SUMMARY_LANGUAGE string with a summary of the scene. It should be in $SUMMARY_LANGUAGE, and it should be less than 30 words. Readers should understand it without looking at the illustration.
-"Illustration": English string with a detailed English description of an illustration for this scene. It must be written in English and in less than 20 words. It should include many details and an artistic style for the image that matches the text.
+"Summary": $SUMMARY_LANGUAGE string with a summary of the scene. It should be in $SUMMARY_LANGUAGE, and it should be less than 50 words. Readers should understand it without looking at the illustration.
+"Illustration": English string with a description of an illustration for this scene in less than 20 words. It should represent the scene, but it should be understandable without the context of the scene.
 Just answer with the JSON object:
 """
 

utils.py

@@ -3,4 +3,8 @@ def is_google_colab():
         import google.colab
         return True
     except:
-        return False
+        return False
+    
+def splitter(n, s):
+    pieces = s.split()
+    return (" ".join(pieces[i:i+n]) for i in range(0, len(pieces), n))