from transformers import WhisperProcessor, WhisperForConditionalGeneration
from peft import PeftModel
import torch

# Load base model and processor
base_model_name = "openai/whisper-large-v3"
processor = WhisperProcessor.from_pretrained(base_model_name)
base_model = WhisperForConditionalGeneration.from_pretrained(base_model_name)

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "WernL/whisper-afrikaans-whisper_training_1756041540")

# Load audio
import librosa
audio, sr = librosa.load("path_to_audio.wav", sr=16000)

# Process
input_features = processor(audio, sampling_rate=16000, return_tensors="pt").input_features
predicted_ids = model.generate(input_features)
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
print(transcription[0])

Alternative: Direct Loading (if supported)

Language: Afrikaans (af)
Base Model: openai/whisper-large-v3
Training Method: LoRA (Low-Rank Adaptation)
Training Steps: 1000
Hardware: gpu-t4
Training Time: N/A hours
LoRA Rank: 8
LoRA Alpha: 32

python

from transformers import pipeline

# This may work if the adapter is properly configured
pipe = pipeline("automatic-speech-recognition", model="WernL/whisper-afrikaans-whisper_training_1756041540")
result = pipe("path_to_audio.wav")
print(result["text"])

Training Configuration

Dataset: common_voice_af_v1
Batch Size: 16
Learning Rate: 1e-05
Max Steps: 1000

Performance

Final training metrics:

WER: 0.089
Loss: 0.177

This model was trained using the Whisper Training App.