A fish shell script that automatically converts a PDF to a plaintext output with aspell's autocorrect.

convert-pdf-to-text

#!/usr/bin/env fish

# Check input arguments
if test (count $argv) -lt 2
    echo "Usage: $argv[0] input.pdf output.txt"
    exit 1
end

set input $argv[1]
set output $argv[2]

# Check if input file exists
if not test -f "$input"
    echo "Error: Input file '$input' not found"
    exit 1
end

# Check dependencies
for cmd in gs tesseract-ocr aspell sed
    if not command -v $cmd &>/dev/null
        echo "Error: $cmd is not installed"
        exit 1
    end
end

echo "Converting PDF to PNG pages..."
gs -dNOPAUSE -dBATCH -sDEVICE=pngalpha -sOutputFile=page_%03d.png $input

echo "Running OCR on each page..."
for img in page_*.png
    set base (basename "$img" .png)  
    echo "  Processing $base"
    tesseract-ocr "$img" "$base"  
end

echo "Compiling pages..."
cat page_*.txt > compiled.txt

echo "Building correction wordlist..."
aspell list < compiled.txt | while read word
set suggestion (echo "$word" | aspell -a | grep "^&" | sed 's/.*: //' | cut -d',' -f1 | string trim)
  if test -n "$suggestion"
    echo "s/\b$word\b/$suggestion/g"
  end
end > auto.sed

echo "Applying autocorrections..."
sed -f auto.sed compiled.txt > $output

# Clean up temporary files
echo "Cleaning up..."
rm -f page_*.png
rm -f page_*.txt
rm -f auto.sed
rm -f compiled.txt

echo "Done! Output: $output"

I'm using Void linux, and the tesseract package is called tesseract-ocr on my system, but may need to be changed to tesseract on yours!