Virtual Keyboard Text Entry Using Speech and Typing Confidence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Text entry on electronic communication devices, particularly touch-screen virtual keyboards, is prone to errors due to the limitations of existing error correction and prediction schemes that do not effectively utilize context-based information from both keyboard key presses and speech recognition.
Innovation Solution
The integration of text prediction and speech recognition systems that assign confidences to candidate words, with speech-recognized candidate words' confidences adjusted based on recency, allowing for the selection of the most likely intended word for faster and more accurate text entry.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If speech recognition is integrated with text prediction to provide more accurate candidate words, then text entry accuracy is improved, but system complexity increases
Solution Approach 1:
The patent combines speech recognition and text prediction systems into a unified architecture where both processes share common components such as the candidate word buffer, confidence scoring mechanism, and selection logic. This merging allows the system to leverage multiple input modalities (typed and spoken words) while avoiding the overhead of completely separate systems, thereby improving accuracy without proportionally increasing complexity
Solution Approach 2:
The system design allows the same candidate word buffer and selection mechanisms to serve multiple functions: storing candidates from text prediction, candidates from speech recognition, and facilitating comparison between both sources. This multi-functionality reduces the need for separate dedicated structures for each input type, managing complexity while enhancing accuracy
2Productivity
If multiple candidate words from both text prediction and speech recognition are processed, then text entry speed is improved, but processing complexity increases
Solution Approach 1:
The system pre-generates candidate words from both text prediction and speech recognition processes before final selection is needed. These candidates are stored in a shared buffer with pre-computed confidence values, allowing the selection process to quickly compare and choose the best match without performing complex real-time analysis, thus speeding up text entry while managing processing complexity
Solution Approach 2:
The system employs confidence scoring as a feedback mechanism to automatically evaluate and rank candidate words from both text prediction and speech recognition. This feedback-driven selection process simplifies the decision-making logic by using quantitative confidence values rather than requiring complex qualitative analysis, improving text entry speed while keeping processing manageable
3Reliability
If speech-recognized candidate words are used with confidence adjustment based on recency, then text entry reliability is improved, but algorithm complexity increases
Solution Approach 1:
The system dynamically adjusts the confidence parameter of speech-recognized candidate words based on the recency of the spoken input relative to the typed input. This parameter change allows the system to weigh speech recognition results more heavily when they are temporally close to the typing action, improving reliability by adapting to the temporal context without requiring complex algorithmic changes beyond simple confidence value modulation
Data Source
AI summary
Typed input is received at a text field from a keyboard, such as a virtual keyboard displayed at an electronic communication device. Text prediction using the typed input can be performed to obtain at least one text-predicted candidate word and an associated confidence value. Speech recognition can be performed on audio input received via a microphone to obtain at least one speech-recognized candidate word and an associated confidence value, which can be adjusted based on the recency of the audio input. A candidate word having a highest confidence value can be selected from the text-predicted and speech-recognized candidate words for display as a suggestion to the user for selection by the user. The suggested candidate word can be displayed on a fret of the virtual keyboard.