signspell

signspell is a Python package for live American Sign Language fingerspelling recognition from a webcam, with CLI, library API, bundled model, and webcam UI.

AI Image Recognition

Visit Website

Overview

signspell is a Python package for live American Sign Language fingerspelling recognition from a webcam. It focuses on the manual alphabet, identifying A–Z in real time and presenting the result through a polished on-screen interface.

The project combines MediaPipe hand tracking with an LSTM model trained on 30-frame keypoint sequences. It ships with a pretrained model, can be used from the command line or as an importable library, and is intended for users who want live webcam-based recognition rather than offline image classification.

Features

ASL fingerspelling recognition

Recognizes the American Sign Language manual alphabet, covering letters A through Z, in real time from webcam input.

Sequence-based recognition pipeline

Uses MediaPipe hand tracking and an LSTM model trained on 30-frame keypoint sequences, according to the project description.

Bundled model and UI

Includes a pretrained model and a polished webcam UI so it can run without wiring up a model yourself first.

CLI controls

Works as a command-line tool with options for camera selection, confidence threshold, mirrored view control, and custom model paths.

Library API

Can be imported as a library, with `signspell.run()` for the full UI and `Recognizer` for programmatic frame-by-frame predictions.

Buffered output handling

Uses a rolling 30-frame buffer and a short stability window to help avoid flicker before a letter is committed.

Use Cases

Live webcam recognition
Run signspell from the terminal to recognize fingerspelled letters from your default camera or a selected webcam index, then adjust the confidence threshold or mirror mode as needed.
Embedded UI usage
Import the package in a Python application and call `signspell.run()` to embed the full webcam UI in your own workflow.
Programmatic frame-by-frame inference
Use the `Recognizer` class with OpenCV frame capture when you want to process frames yourself and handle the predicted letters in code.
Custom model integration
Replace the bundled model with a custom `.h5` file when you already have a compatible model that matches the expected input and output shape.
Interactive ASL demos
Use the package for accessible demos, classroom examples, or prototypes where real-time ASL manual alphabet recognition is the main interaction pattern.

Pros and Cons

Pros

Recognizes the ASL manual alphabet in real time from webcam input.
Provides both CLI and library usage for different workflows.
Ships with a pretrained model and a UI out of the box.
Supports a custom model path for users who want to swap in their own model.
Documents concrete runtime behavior such as camera selection, thresholding, mirroring, and keyboard controls.

Cons

A webcam is required for live recognition.
The source limits supported Python versions to 3.9–3.11.
The project description does not provide setup details beyond installation and basic CLI/library examples.

FAQ

How can I use signspell?

It provides a command-line interface and an importable Python library. The CLI can use the default webcam or a selected camera, and the library exposes `signspell.run()` plus a `Recognizer` class for frame-by-frame predictions.

What are the runtime requirements?

The project description says it requires Python 3.9–3.11 and a webcam for live recognition.

What does it recognize?

The source says it recognizes the American Sign Language manual alphabet (A–Z) in real time and includes a stability window to reduce flicker before a letter is committed.

What license does signspell use?

PyPI lists an MIT License (MIT) for the project.

Can I use my own model?

You can point it at a custom model with `--model path/to/your_model.h5` or `Recognizer(model_path=...)`. The source says the model must accept input shape `(1, 30, 63)` and output 26 class scores.