Systems and methods for real-time neural text-to-speech

Abstract

Embodiments of a production-quality text-to-speech (TTS) system constructed from deep neural networks are described. System embodiments comprise five major building blocks: a segmentation model for locating phoneme boundaries, a grapheme-to-phoneme conversion model, a phoneme duration prediction model, a fundamental frequency prediction model, and an audio synthesis model. For embodiments of the segmentation model, phoneme boundary detection was performed with deep neural networks using Connectionist Temporal Classification (CTC) loss. For embodiments of the audio synthesis model, a variant ofWaveNet was created that requires fewer parameters and trains faster than the original. By using a neural network for each component, system embodiments are simpler and more flexible than traditionalTTS systems, wherein each component requires laborious feature engineering and extensive domain expertise. Inference with system embodiments may be performed faster than real time.

Description

[0001] Cross References to Related Applications [0002] This application was filed on February 24, 2017 under 35USC § 119(e), entitled "SYSTEMS ANDMETHODS FOR REAL-TIME NEURAL TEXT-TO-SPEECH (System and Method for Real-Time Neural Text-to-Speech)" Priority Benefit of U.S. Provisional Patent Application Serial No. 62 / 463,482 (Docket No. 28888-2105P) for Mohammad Shoeybi, Mike Chrzanowski, John Miller, Jonathan Raiman, Andrew Gibiansky, Shubharhrata Sengupta, Gregory Diamos, Sercan Arik, and Adam Coates listed as the inventor. The above patent documents are incorporated herein by reference in their entirety. technical field [0003] The present disclosure relates generally to systems and methods for computer learning that provide improved computer performance, features, and use. More specifically, the present disclosure relates to systems and methods for converting text to speech using deep neural networks. Background technique [0004] Synthesizing artificial human speech...

AI Technical Summary

Problems solved by technology

Because of this complexity, developing new TTS systems can be very labor-intensive and difficult

Furthermore, these systems often work offline and cannot perform text-to-speech conversion in real-time, further limiting their use

Images