EYEDOL/parakeet-tdt-0.6b-yoruba

Parakeet-TDT-0.6B-v2 fine-tuned on Yoruba

This model is a fine-tuned version of nvidia/parakeet-tdt-0.6b-v2 for Yoruba (Yorùbá) automatic speech recognition.

Training details

Parameter	Value
Base model	nvidia/parakeet-tdt-0.6b-v2
Dataset	EYEDOL/naija-voices-yoruba-split_0-1
Train samples	19,374
Val samples	1,022
Epochs	250
Batch size (effective)	32
Learning rate	5e-05
Warmup steps	200
Encoder frozen	True
Final val WER	121.87%

Usage

import nemo.collections.asr as nemo_asr

model = nemo_asr.models.ASRModel.from_pretrained("EYEDOL/parakeet-tdt-0.6b-yoruba")
transcriptions = model.transcribe(["path/to/audio.wav"])
print(transcriptions)

WandB training run

Logged under project parakeet-yoruba-finetune, run parakeet-tdt-0.6b-v2-yoruba.