_______                            _ _                       _____ _      _____ 
 |__   __|                          (_) |                     / ____| |    |_   _|
    | |_ __ __ _ _ __  ___  ___ _ __ _| |__   ___ _ __ ______| |    | |      | |  
    | | '__/ _` | '_ \/ __|/ __| '__| | '_ \ / _ \ '__|______| |    | |      | |  
    | | | | (_| | | | \__ \ (__| |  | | |_) |  __/ |         | |____| |____ _| |_ 
    |_|_|  \__,_|_| |_|___/\___|_|  |_|_.__/ \___|_|          \_____|______|_____|
                                                                                  
                                                                                  

A command-line interface tool for transcribing herbarium label details from images using AWS Bedrock AI models.

Overview

Transcriber CLI is designed to process and transcribe text from herbarium specimen images:

• First Shot: Processes full images to extract label information
• Second Shot: Processes Images once more. (First shot results + Image) for another pass

Prerequisites

• Python 3.x
• AWS account with Bedrock access
• Properly configured AWS credentials: AWS CLI

Installation

1. Clone this repository:

   git clone https://github.com/rherbst123/Transcriber-CLI-V2
   cd Transcriber_CLI
   

2. • Create a virtual enviroment

   python3 -m venv "Whatever you want to call the venv"
   

3. Install required dependencies:

   pip install -r requirements.txt
   

Usage

Run the main script:

python TranscribeCLI.py

The tool will:

1. Ask you to name the run
2. How many Runs to do
3. Choose Model For First run
4. Let Run complete
5. Ask Model for second shot (If chosen)
6. Output a completed .csv file to your desktop with a cost analysis and raw .json files

Validation Features

The tool includes comprehensive validation capabilities that can be configured in the settings menu:

1. Scientific Name Validation

• Validates scientific names against the Global Names Validator on Tropicos
• Flags potential issues with taxonomic nomenclature
• Provides suggested corrections when available

2. Duplicate Records Validation

• Searches the Bryophyte Portal for existing records by catalog number
• Identifies potential duplicate specimens in the database
• Adds detailed duplicate information to CSV output

3. Duplicate Entries Validation

• Searches for existing entries based on collector, collection date, and collection ID
• Helps identify specimens that may already be in the database under different catalog numbers
• Useful for detecting field collection duplicates

All validation features can be enabled/disabled in the configuration menu and run automatically after transcription.

Future Updates

• Scientific Name Validation (Done with Global Names Validator on Tropicos) Global Names
• Search in Portal for Duplicate Catalog numbers
• Search in Portal for Entries on "Collector, record Number and Date"
• Automatic Segmentation for all images done before transcription
• Validation using GBIF, IDigBio and Sybiota
• Enhanced genus/species validation

Supported AI Models

The tool supports multiple AWS Bedrock models:

• Claude 3 Sonnet
• Claude 3.7 Sonnet
• Claude 4 Sonnet
• Claude 4.5 Sonnet
• Claude 4 Opus
• Claude 4.1 Opus
• LLama 3.2 90b
• LLama 4 17b
• Amazon Nova-lite,pro,premier
• Mistral Pixtral Large

More models will be added as they come out.

Output

Transcription results are saved in:

• Desktop/Finished_Transcriptions_"User Entered Run Name"
• Contents Include
  • Segmented Images
  • First and Second shot raw data
  • First and Second shot .csv files

Prompts

The tool uses specialized prompts for herbarium label transcription, located in the Prompts/ directory. The default prompt (Prompt_1.5.3.txt) is designed to extract detailed information from herbarium labels following specific formatting rules.

Customization

• Modify prompts in the Prompts/ directory to adjust transcription behavior
• Add or remove models in the AVAILABLE_MODELS list in each transcriber module (This is updated frequently so you dont have to really)

Created by Riley Herbst, for the Field Museum. With much thanks to the following: Matt Von Konrat, Jeff Gwilliam, Dan Stille