Kasapa-V1
Kasapa-V1 is a fine-tuned version of openai/whisper-small for Automatic Speech Recognition in the Twi language.
Model Description
This model was trained as part of a project to improve accessibility for Twi speakers. It leverages the robust architecture of Whisper-small, adapted specifically for the phonetic nuances of the Twi language.
- Developed by: Willie999
- Model type: Whisper (ASR)
- Language(s) (NLP): Twi (tw)
- License: apache-2.0
- Finetuned from model: openai/whisper-small
Intended Uses & Limitations
Intended Use
This model is intended for transcribing Twi audio into text. It is particularly useful for applications in agriculture, healthcare, and education where Twi is the primary language of communication.
Limitations
Performance may vary based on dialectal differences (e.g., Asante Twi vs. Fante Twi) and background noise levels in the audio recordings.
Training Procedure
Training Hyperparameters
- Learning Rate: 1e-5
- Batch Size: 16
- Steps: 4000
- Optimizer: Adam
- Mixed Precision: FP16
How to Use
from transformers import pipeline
pipe = pipeline("automatic-speech-recognition", model="Willie999/Kasapa-V1")
transcription = pipe("path_to_your_audio.wav")["text"]
print(transcription)