---
language:
- ach
tags:
- text-to-speech
- acholi
- low-resource
- spark-tts
- fine-tuning
---

# Spark-TTS Acholi Fine-tune -- v12

This model is part of a **16-experiment series** fine-tuning Spark-TTS for Acholi (ach) text-to-speech, conducted as part of ongoing Masters research on offline AI dubbing for low-resource African languages.

> Model weights and detailed experimental findings will be published after thesis submission.

## Full Documentation

For the complete experiment log, methodology, and evaluation results across all 16 experiments, see the canonical model:

**[acellam/spark-tts-salt-ach-v16](https://huggingface.co/acellam/spark-tts-salt-ach-v16)**

## About This Experiment

**Version**: v12 -- Speed augmentation 3x + multispeaker SNR filtering

**Language**: Acholi (ach) -- Nilotic (Luo branch), ~1.5M speakers, northern Uganda and South Sudan

**Pipeline**: Whisper ASR -> SALT NLLB-1.3B Translation -> Spark-TTS Synthesis

**Training scripts**: [gitlab.com/mistaguy6/spark-tts-training](https://gitlab.com/mistaguy6/spark-tts-training)