✍️ Twi Punctuation & Capitalization Restorer

Restore commas, sentence-ending periods, and proper capitalization (Title Case, ALL CAPS) in Akan/Twi text.

  • Why this matters for Text-to-Speech (TTS): Commas are routinely omitted when writing Twi, despite their clear presence as pauses in normal spoken Twi. This model compensates for these conventionally ignored commas, enabling natural breath pauses and prosody for TTS engines.
  • Target Use-Cases: Text-to-Speech (TTS) script pre-processing, ASR / Speech-to-Text transcript formatting, and conversational text cleaning.
  • Model: Fine-tuned Ghana-NLP/distilabena-base-v2-akuapem-twi-cased running on quantized INT8 ONNX (~15–28 ms CPU latency).
Examples

Developed by Ghana NLP & Ghana OpenAI. Model available on Hugging Face: ghanaopenai/twi-punct-cap-restoration.