Whisper-Medium-KsponSpeech

The Whisper-medium Model finetunned with KsponSpeech

Model Description

  • Developed by : yw0nam
  • Shared by : yw0nam
  • Model type : ASR
  • License: [apache-2.0]

Uses


processor = WhisperProcessor.from_pretrained("openai/whisper-medium", language="ko", task="transcribe")
model = WhisperForConditionalGeneration.from_pretrained('spow12/whisper-medium-zeroth_korean').cuda()

data, _ = librosa.load(wav_path, sr=16000)
input_features = processor(data, sampling_rate=16000, return_tensors="pt").input_features.cuda()

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

Metrics

Metric result
WER 3.96
CER 1.71
Downloads last month
24
Safetensors
Model size
764M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train mobi/whisper-medium-zeroth_korean