---
library_name: transformers
tags:
  - aphasia
  - text-normalization
  - seq2seq
  - nlp
---

# Model Card for Aphasia Text Normalization

This is a fine-tuned model designed to normalize aphasic speech patterns into standard English, providing better communication capabilities for individuals with speech difficulties.

## Model Details

### Model Description

- **Developed by:** Leif Rogers
- **Shared by:** Leif Rogers
- **Model type:** Seq2Seq Language Model
- **Language(s):** English (EN)
- **License:** Apache 2.0
- **Finetuned from:** T5-Small

The model was fine-tuned on a synthetic dataset generated to mimic aphasic speech patterns and their normalized counterparts. It is intended for applications in assistive technologies to aid individuals with speech impairments.

### Model Sources

- **Repository:** [GitHub Repo](https://github.com/leifsternyc/aphasiamodels)
- **Paper:** Not applicable
- **Demo:** Not available yet

## Uses

### Direct Use

The model can be used directly for text normalization tasks to convert aphasic speech into standard English.

### Downstream Use

Potential downstream uses include integration into assistive communication applications, healthcare tools, or educational resources for speech therapy.

### Out-of-Scope Use

The model is not designed for:
- Speech-to-text conversion
- Non-English languages
- Malicious applications (e.g., creating misleading outputs)

## Bias, Risks, and Limitations

### Bias

The model was trained on synthetic data, which may not represent real-world variations in aphasic speech patterns. It could produce biased outputs for certain dialects or speech patterns.

### Risks

- Overgeneralization of input
- Misinterpretation of ambiguous input phrases

### Recommendations

Users should evaluate the model’s performance in their specific use cases before deployment and provide manual oversight where necessary.

## How to Get Started with the Model

Use the following code to load and use the model:

```python
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

model_name = "leifsternyc/aphasia-t5-normalization"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

# Example usage
input_text = "Want go food need"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))