Downloads · 30 days
17
18% of all-time downloads
AI-for-Education/sw-tz-child-egra-fastconformer-ctc-110m
sw-tz-child-egra-fastconformer-ctc-110m is a automatic speech recognition model from AI-for-Education. Use it when you need speech turned into text. It is set up for nemo. The card lists the license as mit.
modelexp41avg.nemo is our main NVIDIA NeMo automatic speech recognition model for Kiswahili child-speech recordings from Early Grade Reading Assessment (EGRA) tasks in Tanzania. It is designed for letter sound identif…
Downloads · 30 days
17
18% of all-time downloads
All-time downloads
97
Public
Repo size
463 MB
Likes
1
Public
Click a slice to open those files.
.nemo463 MB · 100%
From the Hugging Face model README
model_exp41_avg.nemo is our main NVIDIA NeMo automatic speech recognition model for Kiswahili child-speech recordings from Early Grade Reading Assessment (EGRA) tasks in Tanzania. It is designed for letter sound identification, syllable identification, non-word reading, and passage reading audio.
For public context on the broader Voice AI data collection and annotation workflow, see Fab AI's Voice AI data collection and annotation for early grade reading assessments guide.
| Field | Value |
|---|---|
| Model artifact | model_exp41_avg.nemo |
| Project model identifier | Model ID 41 |
| Task | Automatic speech recognition |
| Language | Kiswahili / Swahili (sw) |
| Target domain | EGRA-style Kiswahili child speech |
| Framework | NVIDIA NeMo |
| NeMo version | 2.3.0 |
| Model class | nemo.collections.asr.models.ctc_bpe_models.EncDecCTCModelBPE |
| Architecture | FastConformer encoder with CTC BPE decoder |
| Parameters | 115,639,795 |
| Input sample rate | 16 kHz |
| Features | 80-dimensional log-mel features, 25 ms window, 10 ms stride |
| Tokenizer | SentencePiece/BPE |
| Vocabulary size | 1,024 BPE tokens |
| Default decoding | Greedy CTC decoding (greedy_batch) |
| Artifact size | 463,144,960 bytes, approximately 442 MiB |
| SHA-256 | 6450926bc1338827ab201b2d9f8f94bcb7a5bd06b6f72690f60ead14a067a7b0 |
This model supports transcription of Kiswahili child speech in education and research workflows, especially EGRA-style oral reading assessments. It can be used as part of a larger assessment pipeline in which transcripts are reviewed, scored, or compared against expected reading prompts.
Appropriate uses include:
The model is not intended for speaker identification, biometric identification, surveillance, clinical transcription, legal transcription, or high-stakes decisions without human review.
We used NVIDIA NeMo 2.3.0 with Python 3.10.12 for local inference and evaluation.
git clone --depth 1 --branch v2.3.0 https://github.com/NVIDIA/NeMo
cd NeMo
conda create -n nemo_env python=3.10.12
conda activate nemo_env
pip install '.[all]'
Load the local .nemo artifact with NeMo:
import torch
from nemo.collections.asr.models import ASRModel
model_path = "model_exp41_avg.nemo"
device = "cuda" if torch.cuda.is_available() else "cpu"
asr_model = ASRModel.restore_from(
restore_path=model_path,
map_location=device,
)
outputs = asr_model.transcribe(["sample.wav"])
for output in outputs:
print(getattr(output, "text", output))
If the model is hosted on Hugging Face, first download the .nemo file:
import torch
from huggingface_hub import hf_hub_download
from nemo.collections.asr.models import ASRModel
repo_id = "AI-for-Education/sw-tz-child-egra-fastconformer-ctc-110m"
model_path = hf_hub_download(repo_id=repo_id, filename="model_exp41_avg.nemo")
device = "cuda" if torch.cuda.is_available() else "cpu"
asr_model = ASRModel.restore_from(
restore_path=model_path,
map_location=device,
)
outputs = asr_model.transcribe(["sample.wav"])
for output in outputs:
print(getattr(output, "text", output))
For the project inference flow, we use NeMo's speech_to_text_eval.py with a JSONL manifest:
{"audio_filepath": "sample.wav", "text": "daima kenya amani tele", "duration": 3.528}
Run:
python3 speech_to_text_eval.py \
model_path=/path/to/model_exp41_avg.nemo \
dataset_manifest=manifest.json \
batch_size=32 \
output_filename=manifest-results.json
The included smoke-test example produces:
{"audio_filepath": "sample1.wav", "text": "daima kenya amani tele", "duration": 3.528, "pred_text": "daima kenya amani tele", "wer": 0.0, "tokens": 4, "ins_rate": 0.0, "del_rate": 0.0, "sub_rate": 0.0}
This example checks that the model restores and runs correctly. It is not a benchmark.
To request timestamps with NeMo's transcription script, pass timestamps=True return_hypotheses=True. With this setting, the output includes char, word, and segment timestamp fields.
Input:
Output:
The basic NeMo transcript output does not add punctuation, capitalization, diarization, or calibrated confidence scores by default. Timestamp output is available through NeMo's timestamp mode.
We evaluated the desktop-runnable Model ID 41 on manually corrected validation and test assessments.
| Evaluation split | Overall WER (%) |
|---|---|
| Validation | 13.0 |
| Test | 16.8 |
| EGRA task | Validation isolated | Validation grid | Test isolated | Test grid |
|---|---|---|---|---|
| Letters | 3.2 | 10.5 | 11.2 | 17.9 |
| Syllables | 21.7 | 19.3 | 22.8 | 23.4 |
| Non-words | 10.1 | 11.3 | 13.5 | 13.6 |
| Passages | - | 8.1 | - | 12.5 |
The strongest results are for passage reading and letter decoding. Syllable decoding remains the most difficult task.
This model was developed as part of a two-phase project. Phase A focused on data collection and annotation for Kiswahili EGRA tasks, and Phase B focused on ASR model development and fine-tuning.
Development included:
Real child-speech data was critical: adult-trained models and pitch-shifted adult speech provided useful baselines, but did not replace real child speech. We observed substantial ASR improvements using as little as 50 hours of child-speech data.
The underlying child-speech assessment data and internal train/validation/test manifests are not included in this repository.
The model uses the following NeMo architecture:
| Component | Configuration |
|---|---|
| Preprocessor | AudioToMelSpectrogramPreprocessor |
| Encoder | ConformerEncoder |
| Encoder layers | 18 |
| Encoder hidden size | 512 |
| Attention heads | 8 |
| Subsampling | Depthwise striding |
| Subsampling factor | 8 |
| Convolution kernel size | 9 |
| Decoder | ConvASRDecoder |
| Decoder output classes | 1,024 |
| Training objective | CTC |
Training settings include AdamW optimization, inverse-square-root learning-rate annealing, warmup ratio 0.03, additive noise augmentation, speed perturbation, and SpecAugment.
This model should be evaluated carefully before use outside the target setting. Known or likely limitations include:
We did not make a separate objective noise-robustness claim because the Phase A dataset did not include noise-level annotations, although many recordings were noisier than initially expected.
The target use case involves children's speech in educational assessment settings. Users should apply strong privacy, consent, data-minimization, and human-review safeguards when handling audio or transcripts. The model should not be used for surveillance or high-stakes decisions about children without appropriate governance, validation, and human oversight.
We developed this model for the Digitisation of EGRA using Voice AI project, with Zevo Tech leading Phase B model development, Fab Inc and Laterite supporting Phase A, and Stellenbosch University collaborating on the evaluation framework.
Public citation language is still being finalized. For now, please cite this model card and the Fab AI Voice AI guide when referencing the model or broader data collection and annotation workflow.