Thousands of Voices for HMM-Based Speech Synthesis–Analysis and Application of TTS Systems Built on Various ASR Corpora

TLDR

Conventional speech synthesis requires large, phonetically balanced studio recordings, limiting voice variety due to high costs, but recent HMM‑based speaker‑adaptive synthesis can robustly use non‑ideal, varied‑condition data, enabling many voices. This study builds high‑quality HMM‑based voices from non‑TTS ASR corpora, demonstrating thousands of voices from datasets such as WSJ, Resource Management, Globalphone, and SPEECON. The authors automatically generate thousands of voices by adapting an average HMM model to each speaker in large ASR corpora and evaluate them perceptually, noting remaining challenges. Experiments confirm that speaker‑adaptive HMM synthesis is robust to non‑ideal data and that perceptual evaluation shows acceptable quality while revealing remaining issues.

Abstract

In conventional speech synthesis, large amounts of phonetically balanced speech data recorded in highly controlled recording studio environments are typically required to build a voice. Although using such data is a straightforward solution for high quality synthesis, the number of voices available will always be limited, because recording costs are high. On the other hand, our recent experiments with HMM-based speech synthesis systems have demonstrated that speaker-adaptive HMM-based speech synthesis (which uses an "average voice model" plus model adaptation) is robust to non-ideal speech data that are recorded under various conditions and with varying microphones, that are not perfectly clean, and/or that lack phonetic balance. This enables us to consider building high-quality voices on "non-TTS" corpora such as ASR corpora. Since ASR corpora generally include a large number of speakers, this leads to the possibility of producing an enormous number of voices automatically. In this paper, we demonstrate the thousands of voices for HMM-based speech synthesis that we have made from several popular ASR corpora such as the Wall Street Journal (WSJ0, WSJ1, and WSJCAM0), Resource Management, Globalphone, and SPEECON databases. We also present the results of associated analysis based on perceptual evaluation, and discuss remaining issues.