Speech Recognition Output Augmentation for Untranscribable Utterances
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems face high error rates due to untranscribable utterances such as slang, local references, and pseudowords, and lack integration with data platforms for further analysis.
Innovation Solution
An end-to-end computing system converts audio streams into intermediate representations, performs diarization, and transforms segments into object-based representations to decipher untranscribable utterances by searching for similar instances within the stream or databases, integrating with data platforms for enhanced contextualization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional speech recognition systems use standard dictionaries for transcription, then processing speed is maintained, but accuracy deteriorates due to untranscribable utterances such as slang, local references, and pseudowords
Solution Approach 1:
The system segments the audio stream into individual speech constructs (phonemes) and processes them through multiple stages: phoneme recognition, dictionary mapping, and fallback mechanisms for untranscribable utterances. This segmentation allows targeted handling of different speech elements without processing the entire stream uniformly.
Solution Approach 2:
The system performs preliminary actions by pre-processing audio streams through voice activity detection, diarization, and phoneme separation before transcription. This preliminary processing prepares the data in advance, enabling more accurate handling of untranscribable utterances when they are encountered.
Solution Approach 3:
The system introduces an intermediary mechanism that acts as a bridge between standard dictionary-based recognition and untranscribable utterances. When a phoneme sequence cannot be mapped to dictionary entries, the system uses intermediate representations and similarity comparisons to find matching instances, rather than directly failing.
2Measurement precision
If the system searches for similar instances to decipher untranscribable utterances, then accuracy improves, but processing time increases
Solution Approach 1:
The system applies partial action by selectively searching for similar instances only when standard dictionary mapping fails for untranscribable utterances. Rather than performing exhaustive similarity searches on every phoneme sequence, the system uses the simpler dictionary approach first and resorts to time-consuming similarity comparisons only when necessary.
Solution Approach 2:
The system changes parameters by adjusting the threshold for what constitutes an untranscribable utterance and modifying similarity comparison criteria. By dynamically adjusting these parameters, the system balances between performing enough similarity searches to improve accuracy and limiting searches to maintain acceptable processing time.
3Loss of information
If the system integrates with data platforms for contextualization, then information completeness improves, but system complexity increases
Solution Approach 1:
The system achieves universality by designing a modular architecture where the data platform integration serves multiple functions: storing phoneme sequences, providing contextual information, and enabling similarity comparisons. This multi-functional design reduces the need for separate specialized components for each function.
Solution Approach 2:
The system uses an intermediary layer that manages communication between the speech recognition components and the data platform. This intermediary handles data formatting, query routing, and result integration, shielding the core recognition logic from platform-specific complexities while maintaining information completeness.
Data Source
AI summary
Computing systems methods, and non-transitory storage media are provided for obtaining an audio stream, converting the audio stream to an intermediate representation, performing diarization on the audio stream, separating the audio stream into individual speech constructs, performing speech recognition on the individual speech constructs by mapping each of the individual speech constructs, or consecutive individual speech constructs, to entries within a dictionary, to generate a transcription of the audio stream, generating an output indicative of the transcription and a result of the diarization, transforming the output into an object-based representation, and performing one or more operations on the object-based representation.