Speech Audio Pre-processing Segmentation via Amplitude Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech-to-text conversion technologies face challenges in accuracy due to issues like environmental noise, dialects, accents, speech impediments, and inadequate preprocessing of speech audio into segments.
Innovation Solution
An apparatus and method that utilize a combination of segmentation techniques, including adaptive peak amplitude and connectionist temporal classification, to identify likely sentence pauses in speech audio, dividing it into segments for improved processing, and employing neural networks for accurate text conversion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If speech audio is divided into equal-length segments without content analysis, then processing efficiency is improved, but speech-to-text conversion accuracy deteriorates due to incomplete phonemes and words being split across segments
Solution Approach 1:
The patent applies segmentation by dividing speech audio into chunks and further into segments, but unlike conventional equal-length segmentation, it uses content-aware segmentation that identifies sentence boundaries based on acoustic features like amplitude thresholds and pause detection. This ensures complete phonemes and words remain within single segments while maintaining efficient processing through systematic division.
Solution Approach 2:
The patent performs preliminary action by analyzing acoustic features and identifying likely sentence pauses before final segmentation occurs. The system pre-processes the speech audio to detect amplitude thresholds, identify pause regions, and mark boundary points, so that when segmentation happens, it aligns with linguistic boundaries rather than arbitrary time intervals.
2Measurement precision
If multiple segmentation techniques are combined to improve accuracy, then speech-to-text conversion accuracy is improved, but device complexity increases
Solution Approach 1:
The patent merges multiple segmentation techniques including amplitude-based pause detection, neural network-based sentence boundary detection, and phoneme boundary analysis. These techniques are combined in a unified framework where their outputs are integrated to identify final segment boundaries, leveraging the strengths of each method while operating within a single processing system.
Solution Approach 2:
The patent uses an intermediary layer that processes outputs from multiple segmentation techniques before final segmentation decisions are made. This intermediary analysis layer reconciles results from different methods, resolves conflicts between them, and produces a unified set of segment boundaries, managing complexity through structured integration rather than direct combination.
3Loss of information
If speech audio is processed as a whole without segmentation, then completeness of speech content is maintained, but processing time and computational resources increase significantly
Solution Approach 1:
The patent applies segmentation to divide the speech audio into manageable chunks and segments based on sentence boundaries. This allows parallel processing of multiple segments simultaneously, reducing overall processing time while maintaining completeness of speech content within each segment. The segmentation ensures no speech information is lost while enabling efficient distributed computation.
Data Source
AI summary
An apparatus includes processor(s) to: divide a speech data set into multiple data chunks that each represent a chunk of speech audio; derive a threshold amplitude based on at least one peak amplitude of the speech audio; designate each data chunk with a peak amplitude below the threshold amplitude a pause data chunk; within a set of temporally consecutive data chunks of the multiple data chunks, identify a longest subset of temporally consecutive pause data chunks; within the set of temporally consecutive data chunks, designate the longest subset of temporally consecutive pause data chunks as a likely sentence pause of a candidate set of likely sentence pauses; based on at least the candidate set, divide the speech data set into multiple data segments that each represent a speech segment of the speech audio; and perform speech-to-text conversion, to identify a sentence spoken in each speech segment.