Tools index

Each tool below is explained in the chapter that uses it. This page collects them in one place for quick reference.

Annotation

• Label Studio: the annotation platform behind AfriAnnotate.
• AfriAnnotate: the AfriPlaybook demo annotation tool.

Speech

• Common Voice: for contributing recordings to Common Voice.
• whisper.cpp: speech recognition that runs on the device.
• Meta MMS: tools for Meta's Massively Multilingual Speech models.

OCR

• Tesseract, TrOCR, Kraken, Calamari, and PaddleOCR.

Model training and inference

• Hugging Face Transformers, PEFT (LoRA and QLoRA), and bitsandbytes.
• llama.cpp, MLC-LLM, and ONNX Runtime Mobile, for running models on phones and small devices.

Metrics

• sacrebleu for chrF and BLEU, jiwer for CER and WER, seqeval for NER, and scikit-learn for classification.

Fonts and script rendering

See also non-Latin script deployment.

• Google Noto: an open font family that covers African scripts.
• SIL fonts: Charis, Doulos, and Andika, for Latin-script text with diacritics.