The text-to-speech in the wild (TITW) database

Traditional Text-to-Speech (TTS) systems rely on studioquality speech recorded in controlled settings. 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 available1 , 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-theart TTS models achieve over 3.0 UTMOS score with TITWEasy, while TITW-Hard remains difficult showing UTMOS below 2.8. Beyond TTS, TITW’s unique design, leveraging a automatic speaker recognition dataset, strengthens ethical efforts to counteract malicious use of TTS models by supporting tasks such as speech deepfake detection.