Sukuma STT Lite
Authors: Zephania and Isack Odero
Compact Sukuma speech recognition for the macron orthography (ā ē ī ō ū). Same task as the quality SKU — Sukuma audio → Sukuma text — with a smaller footprint and higher WER.
Prefer nileagi/nileagi-suk-stt when accuracy matters more than size.
| Quality sibling | nileagi/nileagi-suk-stt |
| Collection | nileagi/nileagi-suk |
Demo
Pitch-shifted demo clip (same file as the quality SKU):
from transformers import pipeline
stt = pipeline(
"automatic-speech-recognition",
model="nileagi/nileagi-suk-stt-lite",
chunk_length_s=30,
ignore_warning=True,
)
print(stt("samples/sample_01.wav")["text"])
Summary
| Task | Automatic speech recognition |
| Language | Sukuma (suk) |
| Direction | Sukuma speech → Sukuma text |
| Orthography | Latin with vowel macrons |
| Training windows | 16,792 · ~98.1 h |
| Test windows | 94 · ~0.5 h |
| Primary metrics | WER, CER |
Evaluation
Same held-out document-group test (94 windows) as the quality SKU:
| Metric | Lite (this repo) | Quality nileagi-suk-stt |
|---|---|---|
| Test WER | 26.1% | 21.3% |
| Test CER | 6.4% | 4.9% |
How to use
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
import torch
repo = "nileagi/nileagi-suk-stt-lite"
device = "cuda:0" if torch.cuda.is_available() else "cpu"
dtype = torch.float16 if device.startswith("cuda") else torch.float32
processor = AutoProcessor.from_pretrained(repo)
model = AutoModelForSpeechSeq2Seq.from_pretrained(repo, torch_dtype=dtype).to(device)
stt = pipeline(
"automatic-speech-recognition",
model=model,
tokenizer=processor.tokenizer,
feature_extractor=processor.feature_extractor,
chunk_length_s=30,
device=device,
ignore_warning=True,
)
print(stt("sukuma.wav")["text"])
Intended use
- Constrained Sukuma ASR where the quality SKU is too heavy
- Same read-aloud setting as the quality model
Out of scope
Anything the quality SKU cannot do: English/Swahili ASR, conversational Sukuma, legal/medical, high-noise audio.
Limitations
Same family as the quality SKU: single-speaker read-aloud setting, small document-group test, macron orthography. Expect more substitutions on names and rare stems than on the quality model.
Related
- Quality STT:
nileagi/nileagi-suk-stt - MT:
nileagi/nileagi-suk-mt - TTS:
nileagi/nileagi-suk-tts
Access
Weights are gated under CC BY-NC-SA 4.0.
- Open Request access at the top of this page and describe your intended use.
- After approval, authenticate (
huggingface-cli loginorHF_TOKEN) and download or load the model.
Attribute NileAGI when you use or redistribute the weights. Adapted weights must stay under CC BY-NC-SA 4.0. Commercial licensing: hi@nileagi.com.
License
Weights: CC BY-NC-SA 4.0.
See LICENSE and NOTICE.md.
Citation
@misc{nileagi-suk-stt-lite-2026,
title = {Sukuma speech-to-text (lite)},
author = {Zephania and Isack Odero},
year = {2026},
howpublished = {Hugging Face},
url = {https://huggingface.co/nileagi/nileagi-suk-stt-lite},
note = {NileAGI}
}