This is a re-packaged, HuggingFace-native version of the CSTR VCTK Corpus, provided as a ready-to-use datasets object (audio decoded via the Audio feature) rather than a loading-script-based dataset.

A companion version resampled to 16kHz is also available: saeedzou/vctk-16khz.

Dataset Summary

This CSTR VCTK Corpus includes around 44 hours of speech data uttered by 110 English speakers with various accents. Each speaker reads out about 400 sentences, which were selected from a newspaper, the rainbow passage, and an elicitation paragraph used for the speech accent archive.

This repository re-hosts the corpus in a native datasets-compatible format (audio stored so it loads directly via the Audio feature at 48kHz), so it can be loaded with:

from datasets import load_dataset

dataset = load_dataset("saeedzou/vctk-48khz")

without requiring a custom loading script.

Supported Tasks

automatic-speech-recognition, speaker-identification: The dataset can be used to train a model for Automatic Speech Recognition (ASR). The model is presented with an audio file and asked to transcribe it to written text. The most common evaluation metric is the word error rate (WER).
text-to-speech, text-to-audio: The dataset can also be used to train a model for Text-To-Speech (TTS).

Languages

English, with a variety of accents from speakers across England, Scotland, Ireland, Wales, the US, Canada, and other regions.

Dataset Structure

Data Instances

A data point comprises the audio, its transcription, and speaker metadata.

{
  'audio':
    {
      'path': 'p225/p225_001_mic1.flac',
      'array': array([0.00485229, 0.00689697, 0.00619507, ..., 0.00811768, 0.00836182, 0.00854492], dtype=float32),
      'sampling_rate': 48000
    },
  'text': 'Please call Stella.',
  'speaker_id': 'p225',
  'age': 23,
  'gender': 'F',
  'accent': 'English',
  'region': 'Southern England',
  'file_id': 'p225_001_mic1'
}

Each audio file is a single-channel FLAC at 48000 Hz.

Data Fields

Each row consists of the following fields:

audio: Audio recording
text: Text transcription of the corresponding audio
speaker_id: Speaker ID
age: Speaker's age
gender: Speaker's gender
accent: Speaker's accent
region: Speaker's region, if annotation exists
file_id: Identifier for the source audio file

Data Splits

The dataset has no predefined splits; a single train split is provided, containing 44,070 examples.

Dataset Creation

Curation Rationale

The original VCTK release required a custom loading script to use with the datasets library. This version repackages the corpus into a native Hugging Face dataset format so it can be loaded directly with load_dataset, with no script execution needed.

Source Data

Initial Data Collection and Normalization

Audio and metadata were sourced from the original CSTR VCTK Corpus (Edinburgh DataShare). No further audio processing was applied for the 48kHz version; files retain their original 48kHz sample rate.

Who are the source language producers?

The speakers who recorded the original VCTK Corpus.

Annotations

Annotation process

Transcriptions and speaker metadata (age, gender, accent, region) are taken from the original VCTK Corpus release.

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

The dataset consists of people who have donated their voice online. You agree not to attempt to determine the identity of speakers in this dataset.

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

Original corpus: Christophe Veaux, Junichi Yamagishi, and Kirsten MacDonald (CSTR, University of Edinburgh). This re-packaged, HuggingFace-native release: saeedzou.

Licensing Information

Public Domain, Creative Commons Attribution 4.0 International Public License (CC-BY-4.0)

Citation Information

@inproceedings{Veaux2017CSTRVC,
    title        = {CSTR VCTK Corpus: English Multi-speaker Corpus for CSTR Voice Cloning Toolkit},
    author       = {Christophe Veaux and Junichi Yamagishi and Kirsten MacDonald},
    year         = 2017
}

Downloads last month: 53

Related repository (16kHz):