CDLI
Collection
This is a collection of models used for the CDLI ASR challenge for atypical speech in Uganda on Ugandan English and Luganda. • 21 items • Updated
Automatic Speech Recognition
NeMo
PyTorch
English
speech
audio
Transducer
TDT
FastConformer
Conformer
NeMo
atypical-speech
dysarthria
Eval Results (legacy)
import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.ASRModel.from_pretrained("KasuleTrevor/cdli-parakeet-11b-en-finetune")
transcriptions = asr_model.transcribe(["file.wav"])
CDLI Parakeet TDT 1.1B English Fine-Tune (lr=5e-5)
This repository contains a NeMo ASR model fine-tuned from
nvidia/parakeet-tdt-1.1b on the gated
cdli/ugandan_english_nonstandard_speech_v1.0 dataset.
This card documents the stronger 1.1B recovery run using a lower learning
rate (5e-5) after the earlier 1e-4 run plateaued early.
Model Details
- Base model:
nvidia/parakeet-tdt-1.1b - Fine-tuning framework: NVIDIA NeMo
- Language: English
- Acoustic model family: FastConformer-TDT / RNNT-BPE
Dataset
- Dataset:
cdli/ugandan_english_nonstandard_speech_v1.0 - License:
cc-by-sa-4.0 - Split sizes used by the source dataset card:
- train:
5176 - validation:
638 - test:
1017
- train:
The evaluation artifacts in this run contain 1016 scored rows.
Training Configuration
- Work root:
/jupyter_kernel/parakeet_cdli_en_5e5 - Base checkpoint:
nvidia/parakeet-tdt-1.1b - Max manifest audio length:
40.0 s - Max training audio length:
30.0 s - Min audio length:
0.2 s - Train batch size:
4 - Eval batch size:
8 - Gradient accumulation steps:
8 - Effective train batch size:
32 - Learning rate:
5e-5 - Weight decay:
1e-3 - Warmup steps:
100 - Scheduler:
CosineAnnealing - Max steps configured:
20000 - Early stopping patience:
10
Evaluation
Evaluation was run on the held-out test split using both raw and normalized
transcript comparison.
Corpus Metrics
- Raw WER:
31.57% - Raw CER:
15.09% - Normalized WER:
21.20% - Normalized CER:
12.56%
Average Utterance Metrics
- Average normalized utterance WER (capped at 1.0):
20.70% - Average normalized utterance CER (capped at 1.0):
12.58%
Files
EN-PARAKEET-TDT-F1tdt-1-1b.nemo: exported NeMo checkpointcheckpoints/: intermediate training checkpointstest_predictions.csvtest_predictions.jsonltest_predictions_scored.csvtest_predictions_scored.jsonltest_predictions_grouped_analysis.csv
Notes
- This
5e-5run improved substantially over the earlier1.1B1e-4run. - Access to the source dataset is gated. Review the dataset terms before requesting access.
- Downloads last month
- 31
Dataset used to train KasuleTrevor/cdli-parakeet-11b-en-finetune
Collection including KasuleTrevor/cdli-parakeet-11b-en-finetune
Evaluation results
- Test WER (raw) on CDLI Ugandan English Non-Standard Speech v1.0test set self-reported31.570
- Test CER (raw) on CDLI Ugandan English Non-Standard Speech v1.0test set self-reported15.090
- Test WER (normalized) on CDLI Ugandan English Non-Standard Speech v1.0test set self-reported21.200
- Test CER (normalized) on CDLI Ugandan English Non-Standard Speech v1.0test set self-reported12.560
# Gated model: Login with a HF token with gated access permission hf auth login