amedcj/whisper-kurmanji

Whisper Kurmanji (ASR)

This is a fine-tuned Whisper model for automatic speech recognition (ASR) on Kurmanji Kurdish.

🗂️ Dataset

Trained on Common Voice Kurmanji.

🛠️ Intended Use

• Transcribing Kurmanji audio into text.
• Fine-tuning for other dialects.

🚀 Usage

from transformers import WhisperProcessor, WhisperForConditionalGeneration
import torch
import torchaudio

processor = WhisperProcessor.from_pretrained("amedcj/whisper-kurmanji")
model = WhisperForConditionalGeneration.from_pretrained("amedcj/whisper-kurmanji").to("cuda")

audio, sr = torchaudio.load("your_audio.wav")
input_features = processor(audio.squeeze(), sampling_rate=16000, return_tensors="pt").input_features.to("cuda")
forced_decoder_ids = processor.get_decoder_prompt_ids(language="kurmanji", task="transcribe")

predicted_ids = model.generate(input_features, forced_decoder_ids=forced_decoder_ids)
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
print(transcription)