Whisper-finetuned-Kikamba
This is a fine-tuned Whisper Small model adapted for Automatic Speech Recognition (ASR) in kamba.
The model was created by training a custom LoRA adapter gamamima/whisper-small-kikamba_lora_adapter on top of openai/whisper-small and subsequently merging the weights for seamless inference.
Model Details
- Base Model:
openai/whisper-small - Architecture: Encoder-Decoder Transformer (Whisper)
- Language: kamba
- Task: Automatic Speech Recognition (ASR)
- Training Method: LoRA (Low-Rank Adaptation) fine-tuning followed by weight merging.
Dataset
gamamima/Kamba-ASR-Data-Subset-484H
Intended Uses & Limitations
Direct Use
The model is intended for transcribing spoken kamba audio into text. It can be used for building localized voice assistants, transcription tools, and accessibility applications.
Limitations & Biases
- Domain Specificity: Performance may vary depending on acoustic conditions, background noise, and specific regional dialects or accents within kamba.
Quick Start (Usage)
from transformers import pipeline
pipe = pipeline("automatic-speech-recognition", model="gamamima/whisper_finetuned_kikamba")
trans = pipe('./audio.wav',kwargs={"task": "transcribe"})