liepa3 / README.md
evaldass's picture
Add official LIEPA-3 project information
f0bf0f3 verified
|
Raw
History Blame Contribute Delete
6.85 kB
metadata
license: cc-by-4.0
language:
  - lt
task_categories:
  - automatic-speech-recognition
  - text-to-speech
pretty_name: LIEPA-3 Lithuanian Speech Corpus
configs:
  - config_name: read
    data_files:
      - split: train
        path: data/read/train-*.parquet
  - config_name: spon
    data_files:
      - split: train
        path: data/spon/train-*.parquet
  - config_name: dial
    data_files:
      - split: train
        path: data/dial/train-*.parquet
  - config_name: phon
    data_files:
      - split: train
        path: data/phon/train-*.parquet
dataset_info:
  - config_name: read
    features:
      - name: id
        dtype: string
      - name: audio
        dtype:
          audio:
            sampling_rate: 44100
      - name: audio_path
        dtype: string
      - name: text
        dtype: string
      - name: normalized_text
        dtype: string
      - name: part
        dtype: string
      - name: split
        dtype: string
      - name: source
        dtype: string
      - name: source_label
        dtype: string
      - name: speech_type
        dtype: string
      - name: speech_type_label
        dtype: string
      - name: speaker_id
        dtype: string
      - name: county
        dtype: string
      - name: region
        dtype: string
      - name: gender
        dtype: string
      - name: age_group
        dtype: string
      - name: textgrid_path
        dtype: string
      - name: utterances_json
        dtype: string
      - name: num_words
        dtype: int32
      - name: audio_nbytes
        dtype: int64
      - name: license
        dtype: string
  - config_name: spon
    features:
      - name: id
        dtype: string
      - name: audio
        dtype:
          audio:
            sampling_rate: 44100
      - name: audio_path
        dtype: string
      - name: text
        dtype: string
      - name: normalized_text
        dtype: string
      - name: part
        dtype: string
      - name: split
        dtype: string
      - name: source
        dtype: string
      - name: source_label
        dtype: string
      - name: speech_type
        dtype: string
      - name: speech_type_label
        dtype: string
      - name: speaker_id
        dtype: string
      - name: county
        dtype: string
      - name: region
        dtype: string
      - name: gender
        dtype: string
      - name: age_group
        dtype: string
      - name: textgrid_path
        dtype: string
      - name: utterances_json
        dtype: string
      - name: num_words
        dtype: int32
      - name: audio_nbytes
        dtype: int64
      - name: license
        dtype: string
  - config_name: dial
    features:
      - name: id
        dtype: string
      - name: audio
        dtype:
          audio:
            sampling_rate: 44100
      - name: audio_path
        dtype: string
      - name: text
        dtype: string
      - name: normalized_text
        dtype: string
      - name: part
        dtype: string
      - name: split
        dtype: string
      - name: source
        dtype: string
      - name: source_label
        dtype: string
      - name: speech_type
        dtype: string
      - name: speech_type_label
        dtype: string
      - name: speaker_id
        dtype: string
      - name: county
        dtype: string
      - name: region
        dtype: string
      - name: gender
        dtype: string
      - name: age_group
        dtype: string
      - name: textgrid_path
        dtype: string
      - name: utterances_json
        dtype: string
      - name: num_words
        dtype: int32
      - name: audio_nbytes
        dtype: int64
      - name: license
        dtype: string
  - config_name: phon
    features:
      - name: id
        dtype: string
      - name: audio
        dtype:
          audio:
            sampling_rate: 44100
      - name: audio_path
        dtype: string
      - name: text
        dtype: string
      - name: normalized_text
        dtype: string
      - name: part
        dtype: string
      - name: split
        dtype: string
      - name: source
        dtype: string
      - name: source_label
        dtype: string
      - name: speech_type
        dtype: string
      - name: speech_type_label
        dtype: string
      - name: speaker_id
        dtype: string
      - name: county
        dtype: string
      - name: region
        dtype: string
      - name: gender
        dtype: string
      - name: age_group
        dtype: string
      - name: textgrid_path
        dtype: string
      - name: utterances_json
        dtype: string
      - name: num_words
        dtype: int32
      - name: audio_nbytes
        dtype: int64
      - name: license
        dtype: string

LIEPA-3 Lithuanian Speech Corpus

This repository repackages the original LIEPA-3 release into Hugging Face Parquet shards with embedded FLAC audio bytes. The original transcriptions are kept as released: normalized lowercase Lithuanian text without punctuation, digits, capitalization, or other symbols.

Recommended use:

  • read: cleanest subset and the default starting point for TTS or ASR.
  • spon: spontaneous/broadcast/media speech; useful for ASR, not a clean TTS default.
  • dial: dialect subset.
  • phon: file-level phonetic metadata with TextGrid paths and utterance lists.

Original LIEPA-3 Project

This Hugging Face dataset is a convenience repackaging of the official LIEPA-3 release, Didžiojo lietuvių kalbos garsyno sukūrimas (LIEPA-3). The official project page states that LIEPA-3 was led by Dr. Gražina Korvel and carried out by Vilnius University with partners Vytautas Magnus University and the Institute of the Lithuanian Language.

The original corpus is available from the CLARIN-LT repository: https://hdl.handle.net/20.500.11821/101

According to the official project description, LIEPA-3 is intended for speech recognition (ASR/STT) applications and research. It contains 10,000 hours of annotated Lithuanian speech in FLAC format (44.1 kHz, 16-bit, mono), with phrase-level annotations for the full corpus and word/phoneme-level annotations for the 500-hour phonetic subset in Praat TextGrid format. The official subcorpora are READ (5,000 h), SPON (4,900 h), DIAL (100 h), and PHON (500 h).

Official project information: https://raštija.lt/liepa-3/apie-projekta-liepa-3/

The project was funded by the Recovery and Resilience Plan "Naujos kartos Lietuva".

Example:

from datasets import Audio, load_dataset

ds = load_dataset("i4tech/liepa3", "read", split="train", streaming=True)
ds = ds.cast_column("audio", Audio(sampling_rate=16000))

Recent datasets versions decode Audio via torchcodec, so install datasets, torch, and torchcodec in training environments. If a pipeline already decodes FLAC itself, use audio["bytes"] directly.