Combining Shinkai tools : from PowerPoint to audio

# /// script
# dependencies = [
#   "requests",
#   "python-pptx",
#   "pytesseract",
#   "pillow",
# ]
# ///

from typing import Optional, List
import os
import requests
import re
import shutil
from pptx import Presentation
from pptx.enum.shapes import MSO_SHAPE_TYPE
import pytesseract
from PIL import Image
import io
from shinkai_local_support import get_home_path
from shinkai_local_tools import shinkai_llm_prompt_processor, eleven_labs_text_to_speech

class CONFIG:
    tesseract_path: str
    generate_audio: str  # "yes" or "no"

class INPUTS:
    file_path: str
    additional_lesson_generation_instructions: Optional[str] = "none"

class OUTPUT:
    content_of_the_presentation: Optional[List[str]]
    status: str
    message: Optional[str]
    lesson_about_the_presentation: Optional[str]
    audio_file: Optional[str]

def extract_text_from_text_frame(text_frame) -> str:
    return ' '.join([run.text for paragraph in text_frame.paragraphs for run in paragraph.runs if run.text])

def extract_text_from_table(table) -> str:
    cell_texts = []
    for row in table.rows:
        for cell in row.cells:
            cell_texts.append(extract_text_from_text_frame(cell.text_frame))
    return ' | '.join(cell_texts)

def extract_text_from_chart(chart) -> str:
    title = extract_text_from_text_frame(chart.chart_title.text_frame) if chart.has_title else ""
    x_axis = extract_text_from_text_frame(chart.category_axis.title.text_frame) if chart.category_axis.has_title else ""
    y_axis = extract_text_from_text_frame(chart.value_axis.title.text_frame) if chart.value_axis.has_title else ""
    return f"Title: {title}, X-Axis: {x_axis}, Y-Axis: {y_axis}"

def extract_text_from_shape(shape) -> List[str]:
    results = []
    
    if shape.has_text_frame:
        text = extract_text_from_text_frame(shape.text_frame)
        if text.strip():
            results.append(f"Text Shape: {text}")
    elif shape.has_table:
        table_text = extract_text_from_table(shape.table)
        if table_text.strip():
            results.append(f"Table: {table_text}")
    elif shape.has_chart:
        chart_text = extract_text_from_chart(shape.chart)
        if chart_text.replace("Title: , X-Axis: , Y-Axis: ", "").strip():
            results.append(f"Chart: {chart_text}")
    elif shape.shape_type == MSO_SHAPE_TYPE.PICTURE:
        try:
            image_data = shape.image.blob
            image = Image.open(io.BytesIO(image_data))
            text = pytesseract.image_to_string(image)
            if text.strip():
                results.append(f"Image Text: {text.strip()}")
        except Exception as e:
            pass
    elif shape.shape_type == MSO_SHAPE_TYPE.GROUP:
        for sub_shape in shape.shapes:
            sub_texts = extract_text_from_shape(sub_shape)
            if sub_texts:
                results.extend(sub_texts)
    
    return results

def clean_special_characters(text: str) -> str:
    text = re.sub(r'\n+', ' ', text)
    text = re.sub(r'#+\s*', ' ', text)
    text = re.sub(r'\*+', ' ', text)
    text = re.sub(r'\s+', ' ', text)
    return text.strip()

def get_base_filename(file_path: str) -> str:
    """Extract the base filename without extension from a path"""
    base_name = os.path.basename(file_path)
    if '?' in base_name:
        base_name = base_name.split('?')[0]
    return os.path.splitext(base_name)[0]

async def _read_presentation(file_path: str, working_file_path: str) -> OUTPUT:
    try:
        if file_path.startswith('http://') or file_path.startswith('https://'):
            response = requests.get(file_path)
            with open(working_file_path, 'wb') as f:
                f.write(response.content)
        else:
            with open(file_path, 'rb') as f:
                with open(working_file_path, 'wb') as w:
                    w.write(f.read())
        o = OUTPUT()
        o.status = "success"
        return o
    except Exception as e:
        o = OUTPUT()
        o.status = "error"
        o.message = f"Failed to read presentation: {str(e)}"
        return o

async def _retrieve_presentation(working_file_path: str) -> OUTPUT:
    if not os.path.exists(working_file_path):
        o = OUTPUT()
        o.status = "error"
        o.message = "No presentation file loaded."
        return o
    try:
        prs = Presentation(working_file_path)
        text_content = []
        for i, slide in enumerate(prs.slides, start=1):
            slide_content = []
            for shape in slide.shapes:
                shape_texts = extract_text_from_shape(shape)
                if shape_texts:
                    slide_content.extend(shape_texts)
            
            if slide_content:
                text_content.append(f"Slide {i}")
                text_content.extend(slide_content)
        
        o = OUTPUT()
        o.content_of_the_presentation = text_content
        o.status = "success"
        return o
    except Exception as e:
        o = OUTPUT()
        o.status = "error"
        o.message = f"Failed to process presentation: {str(e)}"
        return o

def validate_config(config: CONFIG) -> List[str]:
    """
    Validate configuration flags and dependencies.
    Returns a list of error messages if any issues are found.
    """
    errors = []
    # Validate tesseract_path
    if not config.tesseract_path:
        errors.append("Tesseract path is not provided.")
    elif not os.path.isfile(config.tesseract_path):
        errors.append(f"Tesseract path does not point to a file: {config.tesseract_path}")
    elif not os.access(config.tesseract_path, os.X_OK):
        errors.append(f"Tesseract executable is not executable: {config.tesseract_path}")
    
    # Validate generate_audio
    if config.generate_audio.lower() not in ["yes", "no"]:
        errors.append(f"Invalid config value for generate_audio: {config.generate_audio}. Must be 'yes' or 'no'.")
    
    return errors

async def run(config: CONFIG, inputs: INPUTS) -> OUTPUT:
    # Perform configuration validation
    errors = validate_config(config)
    if errors:
        o = OUTPUT()
        o.status = "error"
        o.message = "Configuration errors: " + "; ".join(errors)
        return o

    # Set Tesseract path after validation
    pytesseract.pytesseract.tesseract_cmd = config.tesseract_path

    home_path = await get_home_path()
    working_file_path = os.path.join(home_path, 'working.pptx')

    if not inputs.file_path:
        o = OUTPUT()
        o.status = "error"
        o.message = "No file path provided"
        return o

    read_result = await _read_presentation(inputs.file_path, working_file_path)
    if read_result.status == "error":
        return read_result

    process_result = await _retrieve_presentation(working_file_path)
    if process_result.status == "error":
        return process_result

    slides_content = process_result.content_of_the_presentation or []
    if not slides_content:
        o = OUTPUT()
        o.status = "warning"
        o.message = "No text content found in the presentation"
        o.content_of_the_presentation = []
        return o

    # Step 2: Formulate the prompt for the LLM
    prompt = (
        "<personality>"
        "You are a domain expert teacher extremely good at creating university level educational material."
        "</personality>"
        "<task>"
        "Create the text for an audio lesson of the following content extracted from a powerpoint."
        "</task>"
        "<context>"
        "The content comes from text extraction and OCR from a powerpoint, and is provided shape by shape and slide by slide."
        "Because of this the content form each slide might be a bit disorganised or sparse. You need to take your time step-by-step to reconnect the different pieces of content and understand how they fit together."
        "This content might also contain special characters coming from the text extraction formatting. Make sure to handle these special characters gracefully, they shouldn't appear in the text for an audio lesson."
        "</context>"
        "<lesson instructions>"
        "Do your best to understand in detailed the content of each slide, taking into consideration that it might be a bit disorganised due to the text extraction process."
        "The lesson should be very detailed and precise, and go over absolutely all the content in an organised manner. Be sure to use the entirety of the content for the lesson."
        "The lesson should be clear and engaging."
        "Make sure to introduce it when you move on to a new (sub)topic, so that the listener easily follow along."
        "</lesson instructions>"
        "<additional_instructions>"
        f"{inputs.additional_lesson_generation_instructions}"
        "</additional_instructions>"
        "<format instructions>"
        "Ensure the output is in one continuous paragraph without any special formatting or extra characters. Do not use and \ n formatting, just make it one long endless paragraph. It should be flowing text. You can still include sections in it but present them orally, not with special formatting."
        "The text must flow. Do not use any formatting, but just plain text."
        "</format instructions>"
        "<content>"
        "Here is the content to use for the lesson: "
        f"{' '.join(slides_content)}"
        "</content>"
    )

    # Step 3: Use the LLM to generate the audio lesson text
    llm_response = await shinkai_llm_prompt_processor({"prompt": prompt, "format": "text"})
    generated_text = llm_response['message']
    
    # Step 4: Clean any special characters from the generated text
    cleaned_text = clean_special_characters(generated_text)

    # Step 5: Convert the lesson text to audio if configured to do so
    audio_file_path = None
    if config.generate_audio.lower() == "yes":
        try:
            # Generate the audio with the TTS function
            tts_response = await eleven_labs_text_to_speech({"text": cleaned_text})
            generated_audio_path = tts_response['audio_file']
            
            # Get the base filename from the input file path and create the desired output path
            base_filename = get_base_filename(inputs.file_path)
            desired_audio_path = os.path.join(home_path, f"{base_filename}.mp3")
            
            # Rename the generated audio file to match the presentation name
            if os.path.exists(generated_audio_path):
                # Check if the destination file exists and remove it if it does
                if os.path.exists(desired_audio_path):
                    os.remove(desired_audio_path)
                # Copy the generated file to the desired location with the new name
                shutil.copy2(generated_audio_path, desired_audio_path)
                # Remove the original generated file
                os.remove(generated_audio_path)
                audio_file_path = desired_audio_path
            else:
                audio_file_path = generated_audio_path  # Fallback to the original path if something went wrong
        except Exception as e:
            o = OUTPUT()
            o.status = "error"
            o.message = f"Failed to generate or rename audio: {str(e)}"
            return o

    output = OUTPUT()
    output.content_of_the_presentation = slides_content
    output.lesson_about_the_presentation = cleaned_text
    output.audio_file = audio_file_path
    output.status = "success"
    return output

<task>
Create a tool that do 3 things :
step 1 : extract the text content of a .pptx file presentation.
step 2 : uses the shinkai_llm_prompt_processor tool to generate the text for a lesson about the presentation, using the extracted content.
step 3 : uses the eleven_labs_text_to_speech tool to create an audio file by converting the generated lesson text to audio.
</task>

<input instructions>
The .pptx file will be given as a path in input.
An other input will be optional lesson instructions, and default to 'none'.
Use strings for both.
</input instructions>

<configuration instructions>
2 configurations.
The path to the Tesseract executable.
A 'yes' or 'no' option for audio generation.
Use strings for both.
</configuration instructions>

<instructions for step 1>
Extract the text content of the .pptx file slide by side and shape by shape.
Use Tesseract OCR to extract text from picture shapes.
The executable path of Tesseract will be set in configuration.
</instructions for step 1>

<instructions for step 2>
First, add a check to see if the content extracted is empty or not. If it's empty say so in an error message. If there is content proceed with the lesson text generation.
Then create a strong prompt to generate the best lesson text from the extracted and format it well.
The prompt must explain that the content is coming from text extraction from a .pptx file, slide by slide and shape by shape, and so that a thorough effort should be made to best understand and put together the content of each slide.
The prompt must include a section for optional additional_lesson_instructions, that will be set in inputs by the user.
The prompt must include a format instructions section, saying that because the text is for audio conversion, it should be flowing, there shouldn't be any special formatting, it should be one long flowing paragraph.
The prompt must say to introduce each new(sub)topic, so that the listener can follow along well.
The lesson text should be clear and engaging.
</instructions for step 2>

<intermediary step instructions>
After the text generation, add an intermediary step to remove special characters if any where still present in the generated text. It should remove special character like ##, **, \n and so on.
</intermediary step instructions>

<instructions for step 3>
This audio generation step is optional. It will be set as 'yes' or 'no' in configuration.
Add a step to validate this configuration, it must be either 'yes' or 'no'.
Use the cleaned generated lesson text and the eleven labs tool to convert the file to audio.
Retrieve the name of the original .pptx file, and use this name to rename the audio file.
</instructions for step 3>

<error instructions>
Handle errors gracefully.
Each step should have its own informative error messaging.
</error instructions>

<output instructions>
The output must contain the extracted content, the generated text, the optional audio file, error logs and a status.
</output instructions>

<task>
Create a tool that extracts the text content of a .pptx file presentation.
</task>

<input instructions>
The .pptx file will be given as a path in input.
Use string format.
</input instructions>

<configuration instructions>
1 configuration : the path to the Tesseract executable.
Use string.
</configuration instructions>

<instructions>
Extract the text content of the .pptx file slide by side and shape by shape.
Make sure to include text extraction from all possible shapes with text, including from tables and graphs.
Use Tesseract OCR to extract text from picture shapes.
The executable path of Tesseract will be set in configuration.
If no content was obtained for a shape, remove it from the output. Do not show shapes with empty content.
</instructions>

<error instructions>
Handle errors gracefully.
Each step should have its own informative error messaging.
</error instructions>

<output instructions>
The output must contain the extracted content, error logs and a status.
</output instructions>

# /// script
# dependencies = [
#   "requests",
#   "python-pptx",
#   "pytesseract",
#   "pillow",
# ]
# ///

from typing import Optional, List
import os
import requests
from pptx import Presentation
from pptx.enum.shapes import MSO_SHAPE_TYPE
import pytesseract
from PIL import Image
import io
from shinkai_local_support import get_home_path

class CONFIG:
    tesseract_path: str

class INPUTS:
    file_path: str

class OUTPUT:
    content_of_the_presentation: Optional[List[str]]
    status: str
    message: Optional[str]

def extract_text_from_text_frame(text_frame) -> str:
   return ' '.join([run.text for paragraph in text_frame.paragraphs for run in paragraph.runs if run.text])  # Extracts and joins all text runs from paragraphs in a text frame

def extract_text_from_table(table) -> str:
   cell_texts = []  # List to store text from each cell
   for row in table.rows:
       for cell in row.cells:
           cell_texts.append(extract_text_from_text_frame(cell.text_frame))  # Extract text from each cell's text frame
   return ' | '.join(cell_texts)  # Join all cell texts with pipe separators

def extract_text_from_chart(chart) -> str:
   title = extract_text_from_text_frame(chart.chart_title.text_frame) if chart.has_title else ""  # Get chart title if it exists
   x_axis = extract_text_from_text_frame(chart.category_axis.title.text_frame) if chart.category_axis.has_title else ""  # Get x-axis title if it exists
   y_axis = extract_text_from_text_frame(chart.value_axis.title.text_frame) if chart.value_axis.has_title else ""  # Get y-axis title if it exists
   return f"Title: {title}, X-Axis: {x_axis}, Y-Axis: {y_axis}"  # Format chart data as a string

def extract_text_from_shape(shape) -> List[str]:
    results = []  # Initialize an empty list to store extracted text
    
    # Check if the shape contains a text frame
    if shape.has_text_frame:
        text = extract_text_from_text_frame(shape.text_frame)  # Extract text from the text frame
        if text.strip():  # If text is not empty after stripping whitespace
            results.append(f"Text Shape: {text}")  # Add it to results with a prefix
    
    # Check if the shape contains a table
    elif shape.has_table:
        table_text = extract_text_from_table(shape.table)  # Extract text from the table
        if table_text.strip():  # If table text is not empty
            results.append(f"Table: {table_text}")  # Add it to results with a prefix
    
    # Check if the shape contains a chart
    elif shape.has_chart:
        chart_text = extract_text_from_chart(shape.chart)  # Extract text from the chart
        # Check if chart text has content beyond empty labels
        if chart_text.replace("Title: , X-Axis: , Y-Axis: ", "").strip():
            results.append(f"Chart: {chart_text}")  # Add it to results with a prefix
    
    # Check if the shape is a picture
    elif shape.shape_type == MSO_SHAPE_TYPE.PICTURE:
        try:
            image_data = shape.image.blob  # Get the image data as binary
            image = Image.open(io.BytesIO(image_data))  # Open image from binary data
            text = pytesseract.image_to_string(image)  # Use OCR to extract text from image
            if text.strip():  # If OCR found text
                results.append(f"Image Text: {text.strip()}")  # Add it to results with a prefix
        except Exception as e:
            pass  # Silently ignore any errors in image processing
    
    # Check if the shape is a group of shapes
    elif shape.shape_type == MSO_SHAPE_TYPE.GROUP:
        for sub_shape in shape.shapes:  # Iterate through all shapes in the group
            sub_texts = extract_text_from_shape(sub_shape)  # Recursively process each sub-shape
            if sub_texts:  # If any text was found
                results.extend(sub_texts)  # Add all results to the main list
    
    return results  # Return the list of all extracted text items

async def read_presentation(file_path: str, working_file_path: str) -> OUTPUT:
    try:
        # If the file is from a URL, download it
        if file_path.startswith('http://') or file_path.startswith('https://'):
            response = requests.get(file_path)  # Send HTTP request to get the file
            with open(working_file_path, 'wb') as f:  # Open destination file in binary write mode
                f.write(response.content)  # Write the downloaded content to the file
        # If it's a local file path, copy it to the working location
        else:
            with open(file_path, 'rb') as f:  # Open source file in binary read mode
                with open(working_file_path, 'wb') as w:  # Open destination file in binary write mode
                    w.write(f.read())  # Read the entire source file and write to destination
        
        # Create a success response
        o = OUTPUT()  # Initialize result object
        o.status = "success"  # Set status to indicate successful operation
        return o  # Return the result
    
    except Exception as e:
        # Handle any errors that occur during file operations
        o = OUTPUT()  # Initialize result object
        o.status = "error"  # Set status to indicate error
        o.message = f"Failed to read presentation: {str(e)}"  # Include error details in message
        return o  # Return the error result

async def *retrieve*presentation(working_file_path: str) -> OUTPUT:
    # Check if the file exists
    if not os.path.exists(working_file_path):
        o = OUTPUT()
        o.status = "error"
        o.message = "No presentation file loaded."
        return o
    
    try:
        # Load the presentation file
        prs = Presentation(working_file_path)
        text_content = []
        
        # Iterate through each slide
        for i, slide in enumerate(prs.slides, start=1):
            slide_content = []
            
            # Extract text from all shapes in the slide
            for shape in slide.shapes:
                shape_texts = extract_text_from_shape(shape)
                if shape_texts:
                    slide_content.extend(shape_texts)
            
            # Add slide number and content if there's any text
            if slide_content:
                text_content.append(f"Slide {i}")
                text_content.extend(slide_content)
        
        # Create successful output object with extracted content
        o = OUTPUT()
        o.content_of_the_presentation = text_content
        o.status = "success"
        return o
        
    except Exception as e:
        # Handle any errors during processing
        o = OUTPUT()
        o.status = "error"
        o.message = f"Failed to process presentation: {str(e)}"
        return o

def validate_config(config: CONFIG) -> List[str]:
    """ Validate configuration dependencies.
    Returns a list of error messages if any issues are found.
    """
    errors = []  # Initialize empty list to collect error messages
    
    # Validate tesseract_path
    if not config.tesseract_path:
        # Check if path is empty or None
        errors.append("Tesseract path is not provided.")
    elif not os.path.isfile(config.tesseract_path):
        # Check if path points to an actual file
        errors.append(f"Tesseract path does not point to a file: {config.tesseract_path}")
    elif not os.access(config.tesseract_path, os.X_OK):
        # Check if file has executable permissions
        errors.append(f"Tesseract executable is not executable: {config.tesseract_path}")
    
    return errors  # Return collected error messages

async def run(config: CONFIG, inputs: INPUTS) -> OUTPUT:
    # Perform configuration validation
    errors = validate_config(config)  # Validates the provided configuration
    if errors:
        o = OUTPUT()
        o.status = "error"  # Set error status
        o.message = "Configuration errors: " + "; ".join(errors)  # Combines all errors into a single message
        return o
    
    # Set Tesseract path after validation
    pytesseract.pytesseract.tesseract_cmd = config.tesseract_path  # Configure OCR tool path from config
    
    home_path = await get_home_path()  # Asynchronously get the home directory
    working_file_path = os.path.join(home_path, 'working.pptx')  # Create path for temporary working file
    
    if not inputs.file_path:
        o = OUTPUT()
        o.status = "error"  # Set error status
        o.message = "No file path provided"  # Error if no input file specified
        return o
    
    read_result = await read_presentation(inputs.file_path, working_file_path)  # Read and copy presentation to working file
    if read_result.status == "error":
        return read_result  # Early return if reading fails
    
    process_result = await retrieve_presentation(working_file_path)  # Extract content from the working presentation
    if process_result.status == "error":
        return process_result  # Early return if processing fails
    
    slides_content = process_result.content_of_the_presentation or []  # Get slide content or empty list if None
    
    if not slides_content:
        o = OUTPUT()
        o.status = "warning"  # Set warning status
        o.message = "No text content found in the presentation"  # Warning if presentation has no text
        o.content_of_the_presentation = []
        return o
    
    output = OUTPUT()
    output.content_of_the_presentation = slides_content  # Store extracted content
    output.status = "success"  # Set success status
    return output

<task>
Add a step that uses the shinkai_llm_prompt_processor tool to generate the text for a lesson about the presentation, using the extracted slides content.
Adapt the code. Keep everything else the same.
</task>

<input instructions>
Add a string input for additional lesson instructions to generate the lesson text. It defaults to 'None' if empty.
</input instructions>

<prompt instructions>
Create a strong prompt to generate the best lesson text from the extracted and format it well.
The prompt must explain that the content is coming from text extraction from a .pptx file, slide by slide and shape by shape, and so that a thorough effort should be made to best understand and put together the content of each slide.
The prompt must include a section for optional additional_lesson_instructions, that will be set in inputs by the user.
The prompt must include a format instructions section, saying that because the text is for audio conversion, it should be flowing, there shouldn't be any special formatting, it should be one long flowing paragraph.
The prompt must say to introduce each new(sub)topic, so that the listener can follow along well.
The lesson text should be clear and engaging.
</prompt instructions>

<additional step instructions>
After the lesson text generation, add a step to remove special characters if any are still present in the generated text. It should remove special character like ##, **, \n and so on.
</additional step instructions>

<error instructions>
Handle errors gracefully.
Each step should have its own informative error messaging.
</error instructions>

<output instructions>
The output must contain the extracted content, the generated text, error logs and a status.
</output instructions>

import re
from shinkai_local_tools import shinkai_llm_prompt_processor

class INPUTS:
    file_path: str
    additional_lesson_generation_instructions: Optional[str] = "none"

class OUTPUT:
    content_of_the_presentation: Optional[List[str]]
    status: str
    message: Optional[str]
    lesson_about_the_presentation: Optional[str]

    # Step 2: Formulate the prompt for the LLM
    prompt = (
        "<personality>"
        "You are a domain expert teacher extremely good at creating university level educational material."
        "</personality>"
        "<task>"
        "Create the text for an audio lesson of the following content extracted from a powerpoint."
        "</task>"
        "<context>"
        "The content comes from text extraction and OCR from a powerpoint, and is provided shape by shape and slide by slide."
        "Because of this the content form each slide might be a bit disorganised or sparse. You need to take your time step-by-step to reconnect the different pieces of content and understand how they fit together."
        "This content might also contain special characters coming from the text extraction formatting. Make sure to handle these special characters gracefully, they shouldn't appear in the text for an audio lesson."
        "</context>"
        "<lesson instructions>"
        "Do your best to understand in detailed the content of each slide, taking into consideration that it might be a bit disorganised due to the text extraction process."
        "The lesson should be very detailed and precise, and go over absolutely all the content in an organised manner. Be sure to use the entirety of the content for the lesson."
        "The lesson should be clear and engaging."
        "Make sure to introduce it when you move on to a new (sub)topic, so that the listener can easily follow along."
        "</lesson instructions>"
        "<additional_instructions>"
        f"{inputs.additional_lesson_generation_instructions}"
        "</additional_instructions>"
        "<format instructions>"
        "Ensure the output is in one continuous paragraph without any special formatting or extra characters. Do not use and \ n formatting, just make it one long endless paragraph. It's should be flowing text. You can still include sections in it but present them orally, not with special formatting."
        "The text must flow. Do not use any formatting, but just plain text."
        "</format instructions>"
        "<content>"
        "Here is the content to use for the lesson: "
        f"{' '.join(slides_content)}"
        "</content>"
    )

    # Step 3: Use the LLM to generate the audio lesson text
    llm_response = await shinkai_llm_prompt_processor({"prompt": prompt, "format": "text"})
    generated_text = llm_response['message']

def clean_special_characters(text: str) -> str:
    text = re.sub(r'\n+', ' ', text)
    text = re.sub(r'#+\s*', ' ', text)
    text = re.sub(r'\*+', ' ', text)
    text = re.sub(r'\s+', ' ', text)
    return text.strip()

    # Step 4: Clean any special characters from the generated text
    cleaned_text = clean_special_characters(generated_text)

    output = OUTPUT()
    output.content_of_the_presentation = slides_content
    output.lesson_about_the_presentation = cleaned_text
    output.status = "success"
    return output

<task>
Add an optional step that uses the eleven_labs_text_to_speech tool to create an audio file by converting the generated lesson text to audio.
Adapt the code. Keep everything else the same.
</task>

<configuration instructions>
Add a string configuration for optional generation of the audio, it's either 'yes' or 'no', it defaults to 'no'.
</configuration instructions>

<instructions>
This audio generation step is optional. It will be set as 'yes' or 'no' in configuration.
Add a step to validate this configuration, it must check it's either 'yes' or 'no'.
Use the cleaned generated lesson text and the eleven labs tool to convert the file to audio.
Retrieve the name of the original .pptx file, and use this name to rename the audio file.
</instructions>

<error instructions>
Handle errors gracefully.
Each step should have its own informative error messaging.
</error instructions>

<output instructions>
The output must contain the extracted content, the generated text, the optional audio file, error logs and a status.
</output instructions>

import shutil
from shinkai_local_tools import shinkai_llm_prompt_processor, eleven_labs_text_to_speech

class CONFIG:
    tesseract_path: str
    generate_audio: str  # "yes" or "no"

class OUTPUT:
    content_of_the_presentation: Optional[List[str]]
    status: str
    message: Optional[str]
    lesson_about_the_presentation: Optional[str]
    audio_file: Optional[str]

def get_base_filename(file_path: str) -> str:
    """Extract the base filename without extension from a path"""
    base_name = os.path.basename(file_path)
    if '?' in base_name:
        base_name = base_name.split('?')[0]
    return os.path.splitext(base_name)[0]

    # Validate generate_audio
    if config.generate_audio.lower() not in ["yes", "no"]:
        errors.append(f"Invalid config value for generate_audio: {config.generate_audio}. Must be 'yes' or 'no'.")

    # Step 5: Convert the lesson text to audio if configured to do so
    audio_file_path = None
    if config.generate_audio.lower() == "yes":
        try:
            # Generate the audio with the TTS function
            tts_response = await eleven_labs_text_to_speech({"text": cleaned_text})
            generated_audio_path = tts_response['audio_file']
            
            # Get the base filename from the input file path and create the desired output path
            base_filename = get_base_filename(inputs.file_path)
            desired_audio_path = os.path.join(home_path, f"{base_filename}.mp3")
            
            # Rename the generated audio file to match the presentation name
            if os.path.exists(generated_audio_path):
                # Check if the destination file exists and remove it if it does
                if os.path.exists(desired_audio_path):
                    os.remove(desired_audio_path)
                # Copy the generated file to the desired location with the new name
                shutil.copy2(generated_audio_path, desired_audio_path)
                # Remove the original generated file
                os.remove(generated_audio_path)
                audio_file_path = desired_audio_path
            else:
                audio_file_path = generated_audio_path  # Fallback to the original path if something went wrong
        except Exception as e:
            o = OUTPUT()
            o.status = "error"
            o.message = f"Failed to generate or rename audio: {str(e)}"
            return o

    output = OUTPUT()
    output.content_of_the_presentation = slides_content
    output.lesson_about_the_presentation = cleaned_text
    output.audio_file = audio_file_path
    output.status = "success"
    return output

PPTX Content Extractor With OCR And Audio Lesson Generator

This tool extracts text from a PowerPoint presentation, creates a lesson text about it, and generates an audio file of the lesson.
It extracts content slide by slide and shape by shape.
It uses OCR to extract text from image shapes (using Tesseract).
It generates a detailed lesson text about the .pptx file using a LLM and optional additional instructions.
Optionally, it generates a mp3 audio file of the lesson using ElevenLabs text-to-speech.
This tool requires Tesseract to be installed for OCR. It also requires the tool “ElevenLabs Text to Speech” from the Shinkai AI Store, and an API key for ElevenLabs text-to-speech if choosing to generate audio files.
You can download Tesseract from here : https://github.com/UB-Mannheim/tesseract/wiki. You can get an ElevenLabs API key here : https://elevenlabs.io/developers.

{
  "homePage": "",
  "configurations": {
    "type": "object",
    "properties": {
      "generate_audio": {
        "description": "'yes' or 'no' flag to optionally generate an audio file of the lesson.",
        "type": "string"
      },
      "tesseract_path": {
        "description": "Path to the Tesseract executable (e.g. C:\\Program Files\\Tesseract-OCR\\tesseract.exe).",
        "type": "string"
      }
    },
    "required": [
      "generate_audio",
      "tesseract_path"
    ]
  },
  "parameters": {
    "type": "object",
    "properties": {
      "additional_lesson_generation_instructions": {
        "description": "Optional additional instructions for lesson text generation",
        "type": "string"
      },
      "file_path": {
        "description": "Path or URL of the PowerPoint file (e.g. C:\\Users\\gille\\Desktop\\reef_photogrammetry_presentation.pptx).",
        "type": "string"
      }
    },
    "required": [
      "file_path"
    ]
  },
  "result": {
    "type": "object",
    "properties": {
      "audio_file": {
        "description": "Path to the generated audio file",
        "nullable": true,
        "type": "string"
      },
      "content_of_the_presentation": {
        "description": "Text content extracted from the presentation",
        "items": {
          "type": "string"
        },
        "type": "array"
      },
      "lesson_about_the_presentation": {
        "description": "Text of the generated lesson",
        "nullable": true,
        "type": "string"
      },
      "message": {
        "description": "Error message or additional info",
        "nullable": true,
        "type": "string"
      },
      "status": {
        "description": "The status of the operation",
        "type": "string"
      }
    },
    "required": [
      "status"
    ]
  },
  "sqlTables": [],
  "sqlQueries": [],
  "oauth": [],
  "runner": "any",
  "operating_system": [
    "linux",
    "macos",
    "windows"
  ],
  "tool_set": "",
  "keywords": [
    "pptx",
    "powerpoint",
    "lesson",
    "presentation",
    "audio"
  ],
  "version": "1.0.0"
}