README.md

metadata

dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: transcript
      dtype: string
    - name: source
      dtype: string

Vivoice relabeled using Qwen3-ASR

The original dataset: capleaf/viVoice Sample rate: 44.1 kHz

How-tos: I use the Qwen/Qwen3-ASR-1.7B models then update the label for the sample (audio - text) which have WER (word error rate) < 15%.