Speech Recognition Grapheme Conversion via Unified Decoder
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for verifying speech recognition results are labor-intensive and costly, especially when checking large numbers of words, and require separate decoders for kana-kanji conversion, leading to complex configurations and increased costs.
Innovation Solution
A speech recognition result output device with storage and processing circuitry that acquires a phonetic sequence, converts it into a feature vector, and uses a language model to output graphemes, potentially sharing decoder functions with existing speech recognition engines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a separate decoder for kana-kanji conversion is provided in addition to the existing speech recognition decoder, then the speech recognition result can be obtained, but the device configuration becomes complex and cost increases
Solution Approach 1:
The patent merges the kana-kanji conversion function into the existing speech recognition decoder by adding a conversion unit that operates within the same decoding framework. This allows the decoder to output both phonetic sequences and their corresponding graphemes using a unified structure, eliminating the need for a separate conversion decoder and reducing overall system complexity.
Solution Approach 2:
The speech recognition decoder is enhanced to perform multiple functions: it not only decodes phonetic sequences from acoustic features but also converts these phonetic sequences into graphemes using an integrated conversion unit. This multi-functional approach allows a single decoder to replace what would traditionally require separate components.
2Measurement precision
If manual speech input and result checking is performed to verify speech recognition output, then the verification process can be completed, but the task requires significant effort and cost when checking large numbers of words
Solution Approach 1:
The patent creates a parallel verification path by generating graphemes from phonetic sequences through the integrated conversion unit, which can be directly compared with expected grapheme outputs. This automated copying and comparison process replaces manual speech input and checking, enabling efficient verification of large numbers of words without requiring human effort.
3Productivity
If the language model and decoder are used to convert phonetic sequences to graphemes, then automated verification is enabled, but the system requires additional processing components
Solution Approach 1:
The conversion unit is merged into the existing decoder architecture, sharing computational resources and processing pipelines. This integration allows the system to convert phonetic sequences to graphemes using the same processing infrastructure already in place for speech recognition, avoiding the need for completely separate verification components.
Data Source
AI summary
According to an embodiment, a speech recognition result output device includes a storage and processing circuitry. The storage is configured to store a language model for speech recognition. The processing circuitry is coupled to the storage and configured to acquire a phonetic sequence, convert the phonetic sequence into a phonetic sequence feature vector, convert the phonetic sequence feature vector into graphemes using the language model, and output the graphemes.