Kasanoma ASR

Overview

Kasanoma ASR is a research-focused speech recognition project for African languages and atypical speech. The goal is to enable personalized, code-switched, and edge-deployable ASR systems with a focus on English & Twi.

Key objectives

• Personalized ASR for atypical or disordered speech
• Code-switched models for English and West African languages
• Edge deployment on low-power devices such as Raspberry Pi
• Assistive features for accessibility and disabled users

Project Components

• project-app/: Next.js application, UI components, Firebase integration, and edge-ready frontend
• Dataset access and loading examples for the Kasa speech dataset
• Research and model development resources for code-switched ASR

Project Kasa Dataset

The Project Kasa Dataset contains paired audio and text transcripts for ASR and NLP research. It is built to support code-switched speech recognition for West African language blends.

Dataset access

This dataset is hosted on Hugging Face as a gated repository. To access it:

1. Create a Hugging Face account
2. Request access on the repository page:
   • https://huggingface.co/datasets/Kennethdot/ghana-english-twi-codeswitch-asr
3. Use a Hugging Face access token during download

Loading the dataset

Use the Hugging Face datasets library to load the data once access is granted:

from datasets import load_dataset

# Authenticate with huggingface-cli or set HF_TOKEN
dataset = load_dataset(
    "Kennethdot/Ghana_English-Twi_Code-switching_ASR",
    use_auth_token=True,
)

print(dataset["train"][0])

Expected format

The dataset follows the standard Hugging Face speech format and includes fields such as:

• speaker_id
• audio (48 kHz waveform)
• transcription
• gender

Repository structure

• project-app/: main app folder

  • src/app/: Next.js pages and layout
  • src/components/: reusable UI components and audio recording interface
  • src/firebase/: Firebase config and authentication utilities
  • src/lib/: helper utilities, prompts, and placeholder assets

• notebooks/: exploratory notebooks for setup and Whisper experimentation

• requirements.txt / environment.yml: Python environment dependencies

Getting started

1. Clone the repository
2. Install dependencies for the app and/or Python notebooks
3. Configure Hugging Face access and Firebase credentials as needed
4. Run the Next.js app from project-app/ or explore dataset notebooks

License and citation

The dataset is available under the CC BY-NC 4.0 license for research and educational use. Please cite the Project Kasa Project contributors if you leverage this work in academic or professional settings.