README.md

metadata

license: unknown
language:
  - zh
pretty_name: suisiann
extra_gated_prompt: Access requires agreeing to non-commercial research use only.
extra_gated_fields:
  Name: text
  Affiliation: text
  Intended use: text
  I agree to use this dataset only for non-commercial research purposes: checkbox
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*.parquet
dataset_info:
  features:
    - name: id
      dtype: string
    - name: audio
      dtype: audio
    - name: text
      dtype: string
    - name: tailo
      dtype: string
    - name: source
      dtype: string

suisiann

SuiSiann corpus: text is Han-lo (hanzi-based) Taiwanese script (same language/script as the audio), tailo is Tai-lo romanization. SuiSiann is a TTS corpus; source audio was 48kHz/24-bit and is resampled to 24kHz/16-bit here.

Dataset Description

• Use case: TTS (Text-to-Speech)
• Language: Taigi
• Source: SuiSiann corpus
• Audio format: 24kHz, mono, 16-bit PCM WAV
• Split: this dataset is train only

Dataset Structure

• id: utterance identifier from the source dataset.
• audio: audio clip (24kHz, mono, 16-bit PCM WAV).
• text: text (originally tw in the source dataset), unchanged.
• tailo: text (originally tailo in the source dataset), unchanged.
• source: which recording/category the utterance came from in the original SuiSiann release (e.g. a drama episode title, or a linguistic example category like "日語外來詞kap例句"); not in the arrow shards, joined back in from the original SuiSiann.csv.

Split is encoded by file path (data/train-*.parquet), not a data column, matching the Hugging Face viewer's split selector.

Statistics

lang_name	hours	n_utts	n_chars_in_utts	secs/utt	chars/sec	n_sents	n_chars_in_sents
Taigi	4.734	3467	69817	4.92	4.1	0	0