Streaming Voice Recognition Acoustic Path Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional streaming voice recognition systems face challenges in merging acoustic paths due to historical dependencies, leading to path expansion and decreased decoding speed, which affects accuracy and user experience.
Innovation Solution
The method involves generating and merging acoustic paths with identical last syllables using a streaming attention model, selecting a preset number of paths based on total decoding scores, and reducing the impact of earlier voice history on current voice fragments, thereby improving accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional streaming acoustic models are used for real-time voice recognition, then historical information can be utilized for decoding, but path expansion occurs and decoding speed decreases
Solution Approach 1:
The patent merges acoustic paths that share identical last syllables into a single merged acoustic path. This consolidation reduces the total number of paths that need to be decoded independently, thereby preventing path expansion while preserving the benefits of historical information. The merging operation combines probability distributions and historical states of multiple paths into one, maintaining recognition accuracy while improving decoding speed.
Solution Approach 2:
The patent introduces a new parameter grouping strategy based on last syllable identity. By organizing acoustic paths according to this parameter (last syllable), the system changes how paths are managed and processed. This parameter-based grouping enables efficient merging of paths with identical last syllables, resolving the contradiction between utilizing historical information and maintaining decoding speed.
2Measurement precision
If multiple acoustic paths are maintained for accurate decoding, then recognition accuracy improves, but computational complexity and processing time increase
Solution Approach 1:
By merging acoustic paths with identical last syllables, the patent reduces the number of separate path calculations required. This consolidation maintains the accuracy benefits of multiple paths while significantly reducing computational complexity. The merged path aggregates probability distributions and historical states, eliminating redundant computations across multiple identical paths.
3Manufacturing precision
If all acoustic paths are decoded completely, then decoding accuracy is maximized, but real-time processing capability is compromised
Solution Approach 1:
The patent merges acoustic paths to reduce the total number of paths requiring complete decoding. This merging strategy maintains decoding accuracy by preserving probability distributions and historical states while reducing processing time through fewer independent decoding operations. The consolidation enables real-time processing by eliminating redundant path computations.
Data Source
AI summary
A method, apparatus, device, and computer readable storage medium for recognizing and decoding a voice based on a streaming attention model are provided. The method may include generating a plurality of acoustic paths for decoding the voice using the streaming attention model, and then merging acoustic paths with identical last syllables of the plurality of acoustic paths to obtain a plurality of merged acoustic paths. The method may further include selecting a preset number of acoustic paths from the plurality of merged acoustic paths as retained candidate acoustic paths. Embodiments of the present disclosure present a concept that acoustic score calculating of a current voice fragment is only affected by its last voice fragment and has nothing to do with earlier voice history, and merge acoustic paths with the identical last syllables of the plurality of candidate acoustic paths.