2019/05/22 by Yi Ren, Ren, Yi, Yangjun Ruan +12 · 1 voice · 583 citations
Computer Science · Engineering · Mathematics · #Artificial intelligence #Artificial neural network #Audio and Speech Processing (eess.AS) #Autoregressive model #Computation and Language (cs.CL) #Computer science #Encoder #Engineering #FOS: Computer and information sciences #FOS: Electrical engineering #Inference #Machine Learning (cs.LG) #Mathematics #Parametric statistics #Sound (cs.SD) #Spectrogram #Speech Recognition and Synthesis #Speech and Audio Processing #Speech and dialogue systems #Speech recognition #Speech synthesis #Transformer #cs.CL #cs.LG #cs.SD #eess.AS #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1905.09263
published in arXiv (Cornell University) (Cornell University) · Accepted by NeurIPS2019
openalex publication_date 2019/05/22 · arxiv published 2019/05/22 · arxiv created 2019/11/20 · arxiv updated 2019/11/21 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Neural network based end-to-end text to speech (TTS) has significantly improved the quality of synthesized speech. Prominent methods (e.g., Tacotron 2) usually first generate mel-spectrogram from text, and then synthesize speech from the mel-spectrogram using vocoder such as WaveNet. Compared with traditional concatenative and statistical parametric approaches, neural network based end-to-end models suffer from slow inference speed, and the synthesized speech is usually not robust (i.e., some words are skipped or repeated) and lack of controllability (voice speed or prosody control). In this work, we propose a novel feed-forward network based on Transformer to generate mel-spectrogram in parallel for TTS. Specifically, we extract attention alignments from an encoder-decoder based teacher model for phoneme duration prediction, which is used by a length regulator to expand the source phoneme sequence to match the length of the target mel-spectrogram sequence for parallel mel-spectrogram generation. Experiments on the LJSpeech dataset show that our parallel model matches autoregressive models in terms of speech quality, nearly eliminates the problem of word skipping and repeating in particularly hard cases, and can adjust voice speed smoothly. Most importantly, compared with autoregressive Transformer TTS, our model speeds up mel-spectrogram generation by 270x and the end-to-end speech synthesis by 38x. Therefore, we call our model FastSpeech.