
Closed
Posted
Paid on delivery
I have a run of Lithuanian newspapers printed in Fraktur between 1820 – 1920 and I am already getting fair results by applying Transkribus’ Danish 19th-century Gothic model. I now need a custom model that properly understands both headlines and body text so the character error rate drops to an acceptable level. You will receive a set of high-resolution scanned pages that can serve as ground-truth material. During training we must pay special attention to modernising the historic glyphs: the Fraktur w needs to be normalised to v and the model must reliably output š, č and ž. If you have a proven workflow for creating layout‐ground-truth, running HTR+ experiments, and iterating on confusion pairs, that is exactly what I’m after. Deliverables • A trained Transkribus model (HTR+) covering headlines and body text • A short report summarising training data volume, epochs, and achieved CER/WER on a held-out test set • The project folder (xml/txt) so I can continue refinement later If you believe additional data augmentation or language modelling would further raise accuracy, feel free to propose it—I’m open to suggestions that get results quickly.
Project ID: 40302353
20 proposals
Remote project
Active 9 hours ago
Set your budget and timeframe
Get paid for your work
Outline your proposal
It's free to sign up and bid on jobs
20 freelancers are bidding on average $154 USD for this job

Hello, Your project on training a custom Transkribus model for Lithuanian Fraktur OCR aligns closely with our expertise in AI-driven solutions and machine learning workflows. We understand the nuances of optimizing models for specific needs, such as modernizing historic glyphs and accurately interpreting both headlines and body text to minimize character error rates. At A2 Design, we have proven methodologies for creating layout-ground-truth and executing HTR+ experiments. We have successfully applied similar techniques in diverse projects, supporting our tailored approach to meet specific client requirements. We'd love to collaborate on your project and leverage our experience to refine the model for your Lithuanian newspapers. Let’s discuss how we can bring your vision to life and explore the next steps!
$100 USD in 1 day
8.8
8.8

⭐⭐⭐⭐⭐ Create a Custom HTR+ Model for Lithuanian Newspapers in Fraktur ❇️ Hi My Friend, I hope you are doing well. I've reviewed your project details and see you're looking for a custom HTR+ model for Lithuanian newspapers. You don't need to look any further; Zohaib is here to help you! My team has successfully completed 50+ similar projects focused on historical text recognition. I will create a reliable model that understands both headlines and body text, ensuring a low character error rate. ➡️ Why Me? I can easily build your custom model with my 5 years of experience in handwriting text recognition. My skills include data preprocessing, model training, and layout analysis. I have a strong grip on using Transkribus and enhancing model accuracy. I will ensure modernized historic glyphs are properly handled, boosting your project's success. ➡️ Let's have a quick chat to discuss your project in detail. I would love to show you samples of my previous work. Looking forward to chatting with you! ➡️ Skills & Experience: ✅ HTR+ Model Training ✅ Data Preprocessing ✅ Layout Analysis ✅ Error Rate Reduction ✅ Transkribus Expertise ✅ Confusion Pair Iteration ✅ Historical Text Recognition ✅ Data Augmentation ✅ Report Writing ✅ Project Management ✅ Quality Assurance ✅ Performance Optimization Waiting for your response! Best Regards, Zohaib
$150 USD in 2 days
7.7
7.7

Woah Hello, It sounds like you're looking to enhance the OCR capabilities for Lithuanian newspapers printed in Fraktur from 1820 to 1920. The goal is to create a custom Transkribus model that accurately recognizes both headlines and body text, while also addressing specific characters like the Fraktur w normalization and ensuring reliable output for š, č, and ž. With 7+ years of experience in OCR and natural language processing, I can develop a tailored model that meets your needs. I have a solid workflow for creating layout-ground-truth, conducting HTR+ experiments, and refining confusion pairs. Additionally, I’m open to incorporating data augmentation or language modeling techniques to boost accuracy. I look forward to collaborating on this project and achieving the results you’re aiming for. Best regards, Ivan Mandinski
$35 USD in 3 days
7.4
7.4

Hi there Thanks for posting this exciting project. I checked your project carefully, I think I can complete your project within your needed timeline. I am super professional in PHP, XML, Translation, FileMaker, OCR, Artificial Intelligence, Natural Language Processing, Text Recognition, Data Augmentation Please ping , I am always online here Thanks Efanntyo -.
$30 USD in 5 days
6.7
6.7

Hi There Your project already has a solid starting point with the Danish Gothic model, and I can take it further by building a custom Transkribus HTR+ model tuned for Lithuanian Fraktur across both headlines and body text. I can work on ground-truth preparation, layout-aware training, confusion-pair refinement, and normalization rules like Fraktur w to v while improving reliable output for š, č, and ž. I understand the goal is lowering CER to a practical level, and I can deliver the trained model, test metrics, and the full project files for future refinement. How many fully transcribed ground-truth pages do you currently have available for training and validation? Best Regards Waqas Ahmad
$140 USD in 7 days
6.1
6.1

Hi, I can create a custom Transkribus HTR+ model for your Lithuanian newspapers in Fraktur, designed to handle both headlines and body text accurately. The focus will be on reducing character errors, including normalising historic glyphs like Fraktur *w* to *v* and correctly outputting š, č, and ž. Using your high-resolution scans as ground truth, I’ll train and iteratively refine the model, applying layout-aware transcription and handling common confusion pairs to maximise accuracy. If appropriate, I can also propose data augmentation or language-model adjustments to further improve results. You’ll receive a fully trained HTR+ model, a brief report summarising training details and achieved CER/WER, and the complete project folder (xml/txt) so you can continue refinement later. Best regards, Usman Bashir
$220 USD in 3 days
5.4
5.4

Hi there, Your job post caught my attention because I have extensive experience training custom Transkribus HTR+ models for historical documents, including 19th-century newspapers with Fraktur script. I understand the need to move beyond the Danish Gothic baseline to a model that handles both headlines and body text while normalizing Fraktur 'w' to 'v' and reliably outputting Lithuanian diacritics like š, č, and ž. My approach would involve preparing your ground truth pages with careful layout analysis, training with iterative refinement to reduce character error rates, and documenting the process so you can continue improvements later. Let's discuss your ground truth pages and timeline. Best regards, Mobasher Reza
$140 USD in 3 days
3.2
3.2

Hello, ⭐Who am I?⭐ I am a developer with strong experience in OCR training, text recognition systems, and machine learning workflows for historical documents, including Fraktur and Gothic scripts. I carefully reviewed your project and it aligns well with my experience in training and optimizing OCR/HTR models for historical printed material. I have worked with ground-truth preparation, layout annotation, and iterative training workflows that improve character recognition accuracy while handling historical typography variations. In my opinion, the key to achieving a lower CER for Lithuanian Fraktur newspapers is combining accurate layout-ground-truth with targeted training iterations that address common confusion pairs. To complete this project, I will prepare structured ground-truth data from your scanned pages, train an HTR+ model in Transkribus for both headlines and body text, and implement normalization rules such as mapping the Fraktur w → v while ensuring proper recognition of Lithuanian characters like š, č, and ž. I will also evaluate the model using a held-out test set, optimize training epochs, and provide a report detailing training data size, configuration, and achieved CER/WER. The full project folder (XML/TXT) will also be delivered so you can continue improving the model later. I focus on your satisfaction and update it until as you want with clear communication anytime. Let's make your project successful. Thanks. Manish.
$500 USD in 7 days
3.2
3.2

Hi, I am Matheus, a senior software developer with over 7 years of experience as you can check my profile. I am a senior engineer with over 7 year of experience on PHP, XML, Translation, FileMaker, OCR, Artificial Intelligence, Natural Language Processing, Text Recognition, Data Augmentation. Please visit my profile to view my latest projects, certificates, and work history. Let's connect in chat to discuss more. Thank you, Matheus
$30 USD in 7 days
0.0
0.0

Hello there, It sounds like you’re looking for someone to create a custom OCR model for Lithuanian Fraktur newspapers from 1820 to 1920. I can help with that! With my 5+ years of experience in OCR and Natural Language Processing, I can build a model that effectively recognizes both headlines and body text, addressing the specific challenges with character error rates you mentioned. My approach would involve using the high-resolution scanned pages as a solid foundation for training. I’ll focus on modernizing the historic glyphs, ensuring that the Fraktur w is normalized to v and that characters like š, č, and ž are accurately recognized. I also bring a proven workflow for layout-ground-truth creation and running HTR+ experiments, which will streamline the training process and improve outcomes. I’m also open to suggestions for data augmentation or language modeling to enhance accuracy further. Best regards, shyimaa
$250 USD in 3 days
0.0
0.0

Chicago, United States
Payment method verified
Member since Apr 23, 2019
$10-30 USD
$30-250 USD
$30-250 USD
$2-8 USD / hour
$30-250 USD
$2-8 USD / hour
₹750-1250 INR / hour
$250-750 USD
$15-25 USD / hour
€750-1500 EUR
$2-8 USD / hour
₹600-1500 INR
₹750-1250 INR / hour
$322.56 USD
$10-30 USD
$25-50 USD / hour
$10-30 USD
$36-60 USD / hour
$215.04 USD
$275.52 USD
$83.33 USD
$1500-3000 USD
$35-46 USD
$25-50 USD / hour
$30-250 USD