Parallel Speech Transcription With Segment Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech to text systems are computationally expensive and introduce latency, and splitting conversations into segments leads to a loss of accuracy due to their design limitations.
Innovation Solution
The system splits conversations into smaller chunks for parallel processing using a transformer neural network and a domain-specific correction module to correct inaccuracies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If speech to text systems process entire conversations at once, then transcription accuracy is maintained, but computational cost increases and latency is introduced
Solution Approach 1:
The patent divides the conversation into multiple time-based segments (e.g., 30-second chunks) that can be processed in parallel. Each segment is transcribed independently by multiple transformer models, then the results are combined and corrected. This segmentation enables faster processing while maintaining accuracy through the correction stage.
2Productivity
If speech to text systems split conversations into segments for parallel processing, then processing speed improves, but a layer of accuracy is lost
Solution Approach 1:
The patent implements a correction stage that acts as feedback mechanism. A correction model reviews the transcriptions from multiple segments and modifies them to fix errors. This feedback loop ensures that the parallel processing approach does not permanently sacrifice accuracy, as the correction stage refines the output based on the full conversation context.
Solution Approach 2:
The patent merges the transcriptions from multiple parallel segment processing into a single unified result. The combination process integrates information across segments and applies correction models to ensure accuracy is maintained at the overall conversation level, effectively combining the speed benefits of parallel processing with the accuracy of holistic processing.
3Measurement precision
If legacy speech to text systems process entire conversations, then accuracy is maintained, but resource consumption increases
Solution Approach 1:
By segmenting the conversation into smaller chunks, the system reduces the computational load on each individual transformer model. Multiple smaller models can process segments in parallel using less total computational resources compared to one large model processing the entire conversation, thereby reducing overall energy consumption while maintaining accuracy through the correction stage.
Data Source
AI summary
Methods and systems for transcribing communications are provided. Methods may include receiving a communication. Methods may include splitting the communication into a plurality of communication segments. Each communication segment may include two or more words. Methods may include transcribing each segment included in the plurality of communication segments, in parallel. The transcribing may include using a transformer neural network to transcribe each segment included in the plurality of communication segments. Methods may include generating a transcription from the transcribing. The transcription may be generated by combining the transcription of each of the communication segments into a combined transcription. Methods may include correcting the combined transcription.