Audio Processing Method Combining Recognition Paths
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The N-best decoding algorithm in end-to-end audio recognition has a small receptive field, leading to decreased decoding efficiency and performance due to the need for increased N values to include more recognition results, which slows down the decoding process.
Innovation Solution
An audio processing method that combines identical text sequences from different recognition paths to improve recognition probabilities, enhancing the receptive field without increasing the N value, by performing a combination process on the first sub-block recognition results to determine a second sub-block recognition result, which is then used to obtain the final text recognition result.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the N value is increased to ensure more different effective recognition results are included in N-best information, then the recognition performance is improved, but the decoding speed is slower and decoding efficiency decreases greatly
Solution Approach 1:
The patent merges identical text sequences from different recognition paths by performing a combination process on the first sub-block recognition results. This combines multiple paths that lead to the same text sequence, improving the recognition probability for identical text sequences without increasing the N value, thus maintaining decoding efficiency while improving recognition performance
2Productivity
If the N-best decoding algorithm uses a small receptive field to maintain fast decoding, then the decoding speed is faster, but the recognition performance is affected due to limited effective recognition results
Solution Approach 1:
The patent introduces a new dimension of processing by dividing the audio block into sub-blocks and performing the combination process at the sub-block level. This allows the algorithm to maintain a small receptive field for fast decoding while expanding the effective receptive field through the combination of multiple paths, thereby improving recognition performance without sacrificing decoding speed
Data Source
AI summary
An audio processing method includes obtaining a current to-be-recognized target audio block of a to-be-recognized audio, recognizing text information corresponding to the target audio block to obtain an audio block recognition result, based on the audio block recognition result, determining a first sub-block recognition result corresponding to a current sub-block formed by a starting audio block to the target audio block of the to-be-recognized audio, performing a combination process on identical text sequences of the first sub-block recognition result corresponding to different recognition paths, determining a second sub-block recognition result of the current sub-block based on a result of the combination process, and determining a text recognition result of the to-be-recognized audio based on the second sub-block recognition result. The combination process improves a recognition probability of an identical text sequence matching any one recognition path of the recognition paths corresponding to the identical text sequences.