Text-To-Speech Synthesis In The Wild
This provides a publicly available dataset for noisy-TTS training, addressing a domain-specific limitation in speech synthesis.
The paper tackles the lack of datasets for noisy text-to-speech synthesis by introducing the TTS In the Wild (TITW) dataset, derived from VoxCeleb1, with state-of-the-art models achieving over 3.0 UTMOS on the enhanced version.
Traditional Text-to-Speech (TTS) systems rely on studio-quality speech recorded in controlled settings.a Recently, an effort known as noisy-TTS training has emerged, aiming to utilize in-the-wild data. However, the lack of dedicated datasets has been a significant limitation. We introduce the TTS In the Wild (TITW) dataset, which is publicly available, created through a fully automated pipeline applied to the VoxCeleb1 dataset. It comprises two training sets: TITW-Hard, derived from the transcription, segmentation, and selection of raw VoxCeleb1 data, and TITW-Easy, which incorporates additional enhancement and data selection based on DNSMOS. State-of-the-art TTS models achieve over 3.0 UTMOS score with TITW-Easy, while TITW-Hard remains difficult showing UTMOS below 2.8.