Access this model

Tell us a little about your use, then request access. Downloads unlock after we approve your request.

These weights are shared under CC BY-NC-SA 4.0 for non-commercial use. Please credit NileAGI, keep adaptations under the same licence, and ask us before any commercial product or paid service. Do not use the model to harm people, break the law, or pass outputs off as human speech or writing.

NileAGI Sukuma STT Lite

From NileAGI. Early research release — request access at the top of the page to download the weights.

nileagi/nileagi-suk-stt-lite is the small Sukuma speech-to-text SKU. Same direction and orthography as the quality model, with a lighter footprint and higher WER.

Use nileagi/nileagi-suk-stt when transcription quality matters more than download size.

Sukuma audio only. English / Swahili speech is out of scope.

Collection: nileagi/nileagi-suk

Quick example

One short demo clip ships in samples/sample_01.wav (16 kHz mono). It is pitch-shifted so the raw speaker identity is not exposed (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
Speakers	1 (read-aloud)
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"])

from huggingface_hub import hf_hub_download
wav = hf_hub_download("nileagi/nileagi-suk-stt-lite", "samples/sample_01.wav")
print(stt(wav)["text"])

Intended use

Constrained Sukuma ASR where the quality SKU is too heavy
Same literary read-aloud register as the quality model

Out of scope

Anything the quality SKU cannot do: English/Swahili ASR, conversational Sukuma, legal/medical, high-noise audio. Do not replace the quality SKU when you need the better transcript.

Limitations

Same family as the quality SKU: single speaker, literary register, small document-group test, macron orthography. Expect more substitutions on names and rare stems than on the quality model.

Quality STT: nileagi/nileagi-suk-stt
MT: nileagi/nileagi-suk-mt

Access

Weights are available on request. Fill in the form at the top of this page and click Request access. Once approved, download with a logged-in Hub token (huggingface-cli login or HF_TOKEN).

Research, teaching, and other non-commercial work is fine under CC BY-NC-SA 4.0 — please attribute NileAGI and share adaptations alike. For commercial use, email hi@nileagi.com.

Please don’t use the model to harm people, break the law, or pass outputs off as human speech or writing.

License

Weights: CC BY-NC-SA 4.0.
Training supervision: NOODL.
See LICENSE and NOTICE.md. Commercial use needs a written agreement — email hi@nileagi.com.

Citation

@misc{nileagi-suk-stt-lite-2026,
  title        = {NileAGI Sukuma speech-to-text (lite)},
  author       = {NileAGI},
  year         = {2026},
  howpublished = {Hugging Face},
  url          = {https://huggingface.co/nileagi/nileagi-suk-stt-lite}
}

Downloads last month: 141

Safetensors

Model size

0.2B params

Tensor type

F32

nileagi
/

nileagi-suk-stt-lite