dicksonsarpong9/qwen3-asr-akuapem-twi

Author:

Overview

Qwen3-ASR-Akuapem-Twi is an automatic speech recognition (ASR) model fine-tuned to transcribe Akuapem Twi speech into text. It adapts the Qwen3-ASR-1.7B foundation model for Ghanaian Twi speech recognition using the Akuapem Twi subset of the Ghana Speech datase#

Training data

Trained on the Ghana Speech dataset (audio + text, 42 Ghanaian language subsets), licensed CC BY-NC 4.0.

Intended use & license

Non-commercial use only (CC BY-NC 4.0, inherited from the training data).

How to use

```python
import torch
from qwen_asr import Qwen3ASRModel

model = Qwen3ASRModel.from_pretrained(
    "ghananlpcommunity/qwen3-twi-asr",
    dtype=torch.bfloat16,
    device_map="cuda:0",
)

result = model.transcribe(
    audio="sample.wav"
)

print(result[0].language)
print(result[0].text)

Training details

• Base model / architecture: Qwen3-ASR-1.7B
• Task: Automatic Speech Recognition (ASR)
• Language subset(s): Akuapem Twi
• Training examples: Train: 51,597 samples Validation: 1,053 samples
• Hardware: NVIDIA H200 (Ghana NLP)
• Training duration: ~4 hours
• Epochs: 3
• Batch size: 8
• Gradient accumulation: 8
• Learning rate: 2e-5
• Final training loss: 2.64
• Inference RTF: 0.275

Acknowledgements

Compute resources provided by AI Skills and Compute Africa (AISCA). Trained on the Ghana NLP H200 GPU. Please keep derivatives non-commercial and share improvements back with the Ghana NLP community (ghananlpcommunity).