Voice Bot Identifier Resolution Using Multi-Layer ASR Clarification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated assistants often misrecognize unique personal identifiers such as email addresses, physical addresses, and usernames due to their unpredictable sequences of letters, numbers, and symbols, leading to inefficient and potentially privacy-compromising actions, additional computational resources, and prolonged human-to-computer dialogs.
Innovation Solution
A voice bot utilizes multiple machine learning layers to process ASR speech hypotheses, generating candidate personal identifiers and refining them through prompts and responses to accurately determine the intended unique personal identifier, leveraging transformer and RNN models to enhance accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional ASR is used to process spoken utterances containing unique personal identifiers, then the system is simple and fast, but misrecognition occurs leading to wasted computational resources and prolonged dialogs
Solution Approach 1:
The patent segments the ASR output into multiple hypotheses rather than accepting a single recognition result. Each hypothesis is evaluated with confidence scores, allowing the system to consider multiple possible interpretations of spoken personal identifiers. This segmentation approach resolves the contradiction by maintaining simplicity while improving accuracy through parallel hypothesis evaluation.
Solution Approach 2:
The patent introduces an intermediary component that acts as a bridge between traditional ASR and final recognition. This intermediary evaluates multiple ASR hypotheses, compares them against known personal identifier patterns, and selects the most accurate interpretation. The intermediary resolves the contradiction by adding a layer of intelligence that improves accuracy without requiring complete system redesign.
2Reliability
If ASR misrecognizes personal identifiers, then processing is quick, but erroneous actions are performed and privacy risks increase
Solution Approach 1:
The patent performs preliminary evaluation of ASR hypotheses before final action execution. By pre-evaluating multiple hypotheses against personal identifier patterns and confidence thresholds, the system identifies potential misrecognizations before they lead to erroneous actions. This preliminary action resolves the contradiction by ensuring action accuracy while minimizing additional dialog time through efficient pre-processing.
Solution Approach 2:
The patent implements feedback mechanisms where confidence scores from hypothesis evaluation inform subsequent processing decisions. When confidence is high, the system proceeds quickly; when confidence is low, additional verification steps are triggered. This feedback loop resolves the contradiction by dynamically adjusting processing based on recognition quality, ensuring accuracy while minimizing unnecessary delays.
3Measurement precision
If multiple machine learning layers are used to resolve personal identifiers, then recognition accuracy improves, but computational resources increase
Solution Approach 1:
The patent applies partial machine learning processing only to portions of ASR output that contain personal identifiers rather than processing entire dialog transcripts. By identifying and selectively processing only the relevant segments containing personal identifiers, the system achieves high accuracy where needed while minimizing overall computational resource consumption.
Solution Approach 2:
The patent applies different processing qualities to different parts of the input data. High-quality machine learning models are applied specifically to personal identifier segments, while other portions of the dialog receive standard ASR processing. This local quality approach resolves the contradiction by concentrating computational resources where they provide maximum benefit to accuracy.
Data Source
AI summary
Implementations are directed to causing a voice bot to utilize a plurality of ML layers in resolving unique personal identifier(s) for a human while the voice bot is engaged in a corresponding conversation with the human. The unique personal identifier(s) can include a unique sequence of alphanumeric characters that is personal to the human. In some implementations, ASR speech hypothes(es) corresponding to spoken utterance(s) that include the unique personal identifier(s) can be processed to generate candidate unique personal identifier(s), given alphanumeric character(s) of the candidate unique personal identifier(s) can be selected, and the voice bot can prompt the human with clarification request(s) to clarify the given alphanumeric character(s) until it is predicted to correspond to the an actual unique personal identifier(s) for the human(s). The unique personal identifier(s) can then be utilized in performance of further action(s) by the voice bot and/or other systems.