Logo Lanfrica

Kubihub/Yelmin--Dagbanli-project

Domain:

natural language processing

Record type:

project
Creator:
Kub
Host:
# Yelmin--Dagbanli-project ## Phase 1: Problem Definition & Rationale "Why." This section explains the purpose of the project to non-technical stakeholders or future collaborators. Project Title: Dagbani Language Identification (LID) for Inclusive AI.(Yelmin-Dagbanli-project) The Mission: To build an automated system that recognizes the Dagbani language, acting as the "entry point" for a voice-based chatbot. Linguistic Context: Document that Dagbani is a Gur language spoken in Northern Ghana with tonal features. Note that it is "low-resource" in the digital world. Dataset Source: Formally cite the WAXAL Dataset (2026) and the specific subset (dag_asr). ## Phase 2: Data Exploration & Preprocessing "How." Here, contains the raw material we will be working with. Dataset Statistics: Record the number of audio clips you are using. Audio Specifications: Note the original format (e.g., 48kHz or 44.1kHz) and your decision to resample to 16,000 Hz. Feature Extraction: Explain that you are converting raw sound waves into Mel Spectrograms or MFCCs. Note for your docs: "We chose Mel Spectrograms because they mimic how the human ear perceives sound, making it easier for the Neural Network to identify tonal shifts unique to Dagbani." ## Phase 3: Model Architecture & Training Document the "Brain." This is where you explain the technical choices. Model Selection: We will start with a Simple CNN, and a Random Forest, then we move to fine-tuning a pre-trained model like Wav2Vec2. The Comparison Group: In Language Classification, you need a "distractor" language. Document which one you chose (English) to test if the model can tell them apart. Training Parameters: Document the Learning Rate, Batch Size, and Epochs. ## Phase 4: Evaluation & Ethical Reflection Document the "Results." How well does it actually work? Accuracy Metrics: Use a "Confusion Matrix" to show if the model ever mistakes Dagbani for a similar-sounding language. Ethical Note: Document the diversity of …