Hi-Fi Multi-Speaker English TTS Dataset

By Evelina Bakhturina, Vitaly Lavrukhin, Boris Ginsburg, Yang Zhang. 2021

140 citations recorded by the source corpus.

This paper introduces a new multi-speaker English dataset for training text-to-speech models. The dataset is based on LibriVox audiobooks and Project Gutenberg texts, both in the public domain. The new dataset contains about 292 hours of speech from 10 speakers with at least 17 hours per speaker sampled at 44.1 kHz. To select speech samples with high quality, we considered audio recordings with a signal bandwidth of at least 13 kHz and a signal-to-noise ratio (SNR) of at least 32 dB. The dataset is publicly released at http://www.openslr.org/109/ .

Subfields: Artificial Intelligence, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Signal Processing

Citations in the corpus

Guided-TTS: A Diffusion Model for Text-to-Speech via Classifier Guidance (2021, methodology): “Two cross-shaped masks (LJSpeech, Hi-Fi TTS male) and one binarized MNIST (LeCun & Cortes, 2010) mask (Hi-Fi TTS female) are used for masking.”
NANSY++: Unified Voice Synthesis with Neural Analysis and Synthesis (2022, methodology): “More details can be found at Appendix B.2 Data efficiency To evaluate data efficiency, we used a single speaker (reader id: 8051) from the Hi-Fi multi-speaker english TTS dataset (Bakhturina et al., 2021).”
WavThruVec: Latent speech representation as intermediate features for neural speech synthesis (2022, methodology): “We additionally use Hi-Fi TTS [31] and VCTK [32] datasets.”
A Survey on Recent Deep Learning-driven Singing Voice Synthesis Systems (2021, background): “…(TTS) is a generation task with similar characteristics and complexity; however, an ample amount of free, open datasets is available for TTS [30], [31], which allows the ﬁeld to progress rapidly and see a wide array of successful architectures and readily available commercial products.”
Distribution Augmentation for Low-Resource Expressive Text-To-Speech (2022, unknown): “• Two high-quality English voices selected from publicly available HiFi TTS dataset [24], one male (D 2 ) and one female (D 2 ).”
Better speech synthesis through scaling (2023, methodology): “, 2019) and HiFiTTS(Bakhturina et al., 2021) datasets, which combined contain 896 hours of transcribed speech.”
DRED: Deep REDundancy Coding of Speech Using a Rate-Distortion-Optimized Variational Autoencoder (2022, methodology): “Both the RDO-VAE and the FARGAN vocoder are trained independently on 205 hours of 16-kHz speech from a combination of TTS datasets [27], [28], [29], [30], [31], [32], [33], [34], [35] including more than 900 speakers in 34 languages and dialects.”