---
dataset_info:
  features:
  - name: client_id
    dtype: string
  - name: path
    dtype: string
  - name: audio
    dtype:
      audio:
        sampling_rate: 16000
  - name: sentence
    dtype: string
  - name: up_votes
    dtype: int64
  - name: down_votes
    dtype: int64
  - name: age
    dtype: string
  - name: gender
    dtype: string
  - name: accent
    dtype: string
  - name: locale
    dtype: string
  - name: segment
    dtype: string
  - name: variant
    dtype: string
  - name: original
    dtype:
      audio:
        sampling_rate: 16000
  splits:
  - name: train
    num_bytes: 1492901277.74
    num_examples: 7101
  - name: validation
    num_bytes: 1078546492.24
    num_examples: 4976
  - name: test
    num_bytes: 1189633988.24
    num_examples: 4976
  download_size: 3109373970
  dataset_size: 3761081758.2200003
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
license: cc0-1.0
language:
  - zh
language_bcp47:
  - zh-TW
pretty_name: Common Voice Corpus 16.1 Clean (zh-TW)
source_datasets:
  - mozilla-foundation/common_voice_16_1
---

This dataset is derived from [mozilla-foundation/common_voice_16_1](https://huggingface.co/datasets/mozilla-foundation/common_voice_16_1), with a clean (denoised) audio column using [MP-SENet](https://huggingface.co/JacobLinCool/MP-SENet-DNS). The original "noisy" audio is stored in the "original" column.