Text-to-Speech Phoneme Processing with Concurrent RNN Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text-to-speech (TTS) technologies using recurrent neural networks (RNNs) for autoregressive processing of phoneme audio features are inefficient due to iterative output processing, leading to slow audio synthesis speeds.
Innovation Solution
A method involving concurrent processing of adjacent phonemes using a target RNN, where feature extraction is performed on phonemes to generate audio features, and speech data is obtained simultaneously for each phoneme, decoupling the processing timeline and reducing computational overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If autoregressive processing is used for phoneme audio features in TTS, then speech data prediction is achieved, but audio synthesis speed decreases
Solution Approach 1:
The patent segments the processing of adjacent phonemes into independent parallel tasks. Instead of sequentially processing phonemes through the RNN (where each phoneme's output depends on the previous one), the method divides adjacent phonemes into separate processing streams that can be handled concurrently, breaking the sequential dependency chain while maintaining phoneme-level accuracy
Solution Approach 2:
The patent performs preliminary feature extraction on multiple adjacent phonemes before RNN processing. By extracting audio features from several phonemes in advance and preparing them for concurrent RNN processing, the system reduces the overall processing time while ensuring that each phoneme receives the necessary contextual information for accurate speech data generation
2Reliability
If iterative output processing is used in RNN, then autoregressive speech prediction is achieved, but computational overhead increases
Solution Approach 1:
The patent merges the processing of multiple adjacent phonemes into a single concurrent RNN operation. Instead of iteratively processing each phoneme through separate RNN passes, the method combines multiple phoneme processing tasks into parallel RNN operations that share computational resources, reducing overall computational overhead while maintaining prediction reliability
Solution Approach 2:
The patent creates multiple copies of the RNN processing path for adjacent phonemes. By instantiating parallel RNN processing streams for different phonemes simultaneously, the system eliminates iterative sequential processing while ensuring each phoneme receives dedicated computational attention, thereby reducing computational overhead without sacrificing prediction accuracy
Data Source
AI summary
The present disclosure relates to text data processing methods and apparatuses. One example method includes obtaining target text, where a phoneme of the target text includes a first phoneme and a second phoneme that are adjacent to each other. Feature extraction is performed on the first phoneme and the second phoneme to obtain a first audio feature of the first phoneme and a second audio feature of the second phoneme. By using a target recurrent neural network (RNN) and based on the first audio feature, first speech data corresponding to the first phoneme is obtained. By using the target RNN and based on the second audio feature, second speech data corresponding to the second phoneme is obtained. By using a vocoder and based on the first speech data and the second speech data, audio corresponding to the first phoneme and audio corresponding to the second phoneme are obtained.