Whisper Small is a fine-tuned version of openai/whisper-small for automatic speech recognition (ASR) in Akan using an English tokenizer.
The model aims to improve speech recognition capabilities for low-resource African languages, particularly AKan, which is widely spoken in Ghana but underrepresented in modern speech technologies.

Model Details

Model Description

This model was developed to transcribe spoken Twi audio into text using the Whisper sequence-to-sequence speech recognition architecture. By fine-tuning Whisper on Twi speech data, the model adapts the multilingual speech representations of the base model to better capture phonetic and linguistic patterns specific to Twi.

Developed by: Tiffany Degbotse
Funded by [optional]: Academic research / personal project
Shared by [optional]: Tiffany Degbotse
Model type: Automatic Speech Recognition (Speech-to-Text)
Language(s): Twi (Akan)
License: Apache-2.0
Finetuned from model: openai/whisper-small

Model Sources

Repository: https://huggingface.co/tiffany101/whisper-small

Uses

Direct Use

This model can be used for:

Speech-to-text transcription of Akan audio
Voice interfaces supporting Ghanaian languages
Accessibility tools for Twi speakers
Linguistic research involving Akan speech

Downstream Use

The model may also be used as a base model for:

Fine-tuning on additional Akan speech datasets
Multilingual ASR systems including Ghanaian languages
Voice assistants and conversational AI applications

Out-of-Scope Use

The model may perform poorly when:

Audio quality is extremely noisy
Speech contains heavy code-switching between languages
Speakers use dialects not represented in the training data

Bias, Risks, and Limitations

Like many speech recognition systems trained on limited datasets, the model may exhibit biases toward accents or speech styles present in the training data. Performance may vary depending on speaker accent, recording quality, and domain differences between training and test audio.

Recommendations

Users should evaluate the model on their own datasets before deployment and remain aware of potential limitations when applying the model to different domains.

How to Get Started with the Model

python

from transformers import pipeline

pipe = pipeline(
    "automatic-speech-recognition",
    model="tiffany101/whisper-small"
)

pipe("audio.wav")