Welcome to nkululeko’s documentation!
This documentation contains installation, usage, INI file format, and tutorials of Nkululeko, Machine Learning Speaker Characteristics. The program is intended for novice people interested in speaker characteristics detection (e.g., emotion, age, and gender) without proficient in (Python) programming language. Main features of Nkululeko are:
Finding good combinations of several variables, e.g., acoustic features and models (classifier or regressor), feature standardization, augmentation, etc., for speaker characteristics detection,
Characteristics of the database, such as distribution of gender, age, emotion, duration, data size, and so on with their visualization,
Inference of speaker characteristics from a given audio file or streaming audio (can be said also as “weak” labeling for semi-supervised learning).
Altogether, this make Nkululeko as a good teaching/learning tool for speaker characteristics detection by machine learning.
The examples only covers some important features of Nkululeko. For more details, please refer to the Nkululeko Github page and Felix’s web page.
There is also a deepwiki available. You can directly ask your question there (Nkululeko Github also can be used with Copilot).
How-to Guides
Visualization
Tutorials
- Nkululeko Workflow
- Hello world [AUDFORMAT]
- Hello World [CSV]
- Emotion Prediction with Emotion2vec
- Comparing classifiers and features
- Feature Scaling in Nkululeko
- Data Balancing in Nkululeko
- Feature Correlation Plots (regplot)
- Activation Functions in Neural Network Models
- Overview
- What are Activation Functions?
- Available Activation Functions
- Configuration
- Practical Examples
- CNN Models with List Layer Format
- Choosing the Right Activation Function
- Performance Comparison Example
- Testing Your Configuration
- Advanced Tips
- Troubleshooting
- Complete Working Example
- Summary
- References
- See Also
- Text Processing: Transcribe, Translate, and Classify
- How to Align Databases
- Using Uncertainty in Predictions
- Comparing Classifiers, Features, and Databases
- Linguistic Features with BERT
- Train/Dev/Test Splits
- Predicting Speaker ID
- Finetuning Transformer Models
- Using Split Train and Test Data
- How to Split Your Data
- Bundle: Export and Inference
- Equal Error Rate (EER) Implementation for Nkululeko
Modules
- nkululeko.nkululeko
- nkululeko.explore
- nkululeko.augment
- nkululeko.resample
- nkululeko.segment
- nkululeko.optim
- Hyperparameter Optimization Module
- Quick Start
- Optimization Approaches
- Real-World Examples with Polish Emotional Speech Dataset
- Choosing the Right Approach
- Configuration Parameters
- Parameter Specification
- Model-Specific Parameters
- Complete Examples
- Understanding Results
- Advanced Features
- Best Practices
- Troubleshooting
- Performance Tips
- Integration with Nkululeko Workflow
- Summary
- nkululeko.predict
- nkululeko.multidb
- nkululeko.ensemble
- nkululeko.flags
API Reference
- nkululeko package
- Subpackages
- Submodules
- nkululeko.aug_train module
- nkululeko.augment module
- nkululeko.cacheddataset module
- nkululeko.constants module
- nkululeko.demo_predictor module
- nkululeko.experiment module
- nkululeko.explore module
- nkululeko.export module
- nkululeko.feature_extractor module
- nkululeko.file_checker module
- nkululeko.filter_data module
- nkululeko.glob_conf module
- nkululeko.modelrunner module
- nkululeko.multidb module
- nkululeko.nkululeko module
- nkululeko.plots module
- nkululeko.predict module
- nkululeko.reporter module
- nkululeko.resample module
- nkululeko.result module
- nkululeko.runmanager module
- nkululeko.scaler module
- nkululeko.segment module
- nkululeko.syllable_nuclei module
- nkululeko.testing_predictor module
- Module contents