AfriSpeech/african-sentiment-id

Universal African Sentiment Classifier

Sentiment classification (positive/ negative) for African languages, trained on universal G2P orthography so it works across all 693 languages without per-language conditioning.

Models

• model.bin / model.ftz — FastText supervised classifier.
• onnx/ — Fine-tuned BERT-tiny exported to ONNX.

Usage

# FastText
import fasttext
model = fasttext.load_model("model.bin")
model.predict("a universal g2p sentence")

# ONNX transformer (onnxruntime + tokenizer)
import onnxruntime, transformers
sess = onnxruntime.InferenceSession("onnx/model.onnx")
tok = transformers.AutoTokenizer.from_pretrained("onnx/")