End-to-End Text-to-Speech Neural Network Eliminates HMM Aligners
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text-to-speech systems are inefficient in generating speech from text, requiring complex components and hand-engineered linguistic features, which increases computational resources and reduces convergence speed.
Innovation Solution
A system implemented as computer programs on one or more computers that uses a sequence-to-sequence recurrent neural network to convert text to speech, generating a spectrogram of a verbal utterance without requiring Hidden Markov Model (HMM) aligners, thus reducing model size, training time, and inference time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If complex components and hand-engineered linguistic features are used in text-to-speech systems, then speech quality can be maintained, but computational resources increase and convergence speed reduces
Solution Approach 1:
The patent extracts and removes complex components such as Hidden Markov Model (HMM) aligners and hand-engineered linguistic features from the text-to-speech system. By eliminating these unnecessary components, the system achieves faster convergence and reduced computational resource requirements while maintaining speech generation quality through the sequence-to-sequence recurrent neural network architecture.
2Reliability
If complex components and hand-engineered linguistic features are used in text-to-speech systems, then speech quality can be maintained, but training time and inference time increase
Solution Approach 1:
The patent removes time-consuming components including HMM aligners and complex hand-engineered features from the system architecture. This extraction eliminates unnecessary processing steps during both training and inference phases, significantly reducing training time and inference time while the sequence-to-sequence recurrent neural network maintains effective speech generation capabilities.
3Reliability
If traditional text-to-speech systems with complex components are used, then comprehensive speech generation is achieved, but model size increases
Solution Approach 1:
The patent extracts and eliminates redundant components such as HMM aligners and extensive hand-engineered linguistic features from the system. This removal reduces the overall model size and computational footprint while the sequence-to-sequence recurrent neural network architecture maintains comprehensive speech generation capabilities through its end-to-end learning approach.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating speech from text. One of the systems includes one or more computers and one or more storage devices storing instructions that when executed by one or more computers cause the one or more computers to implement: a sequence-to-sequence recurrent neural network configured to: receive a sequence of characters in a particular natural language, and process the sequence of characters to generate a spectrogram of a verbal utterance of the sequence of characters in the particular natural language; and a subsystem configured to: receive the sequence of characters in the particular natural language, and provide the sequence of characters as input to the sequence-to-sequence recurrent neural network to obtain as output the spectrogram of the verbal utterance of the sequence of characters in the particular natural language.