Speech Recognition Using Compressed Character Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition technologies face challenges in efficiently converting audio frames into character string data while managing data size and processing speed.
Innovation Solution
A speech recognition apparatus that employs two models: a first model to convert audio frames into compressed character string data, and a second model, learned as a decoder from an autoencoder, to convert the compressed data back into character string data, thereby reducing data size and improving processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If character string data is directly generated from audio frames, then speech recognition accuracy is maintained, but data size becomes large and processing efficiency decreases
Solution Approach 1:
The patent introduces compressed character string data as an intermediary between audio frames and final character string data. The first model converts audio frames to compressed character string data, which serves as a compact intermediate representation. The second model then converts this compressed data back to full character string data, achieving both data size reduction and maintained processing efficiency.
Solution Approach 2:
The patent changes the parameter of data representation by introducing a compressed format. Instead of using full character string data directly, the system transforms the data into a compressed representation that occupies less space while preserving the essential information needed for speech recognition, thereby improving processing efficiency without sacrificing accuracy.
2Speed
If compressed character string data is used, then data size is reduced and processing speed is improved, but the system requires additional model components
Solution Approach 1:
The patent segments the speech recognition task into two distinct stages handled by separate models. The first model is responsible for converting audio frames to compressed character string data, while the second model handles the conversion from compressed data to final character string data. This segmentation allows each model to be optimized for its specific function, improving overall processing speed despite the added complexity.
3Productivity
If a two-model system is implemented, then processing efficiency is improved, but the system requires more training data and learning resources
Solution Approach 1:
The patent applies preliminary action by first training the second model (decoder) using autoencoder learning before training the first model. The second model is pre-trained to effectively convert compressed character string data back to full character string data. This preliminary training ensures that when the first model generates compressed data, the second model is already optimized to reconstruct it accurately, reducing the overall training data requirements and improving processing efficiency.
Data Source
AI summary
A speech recognition apparatus (2000) includes a first model (10) and a second model (20). The first model (10) is learned by training data with an audio frame as input data, and with, as correct answer data, compressed character string data acquired by encoding character string data represented by the audio frame. The second model (20) is a learned decoder (44) acquired by learning an autoencoder (40) being constituted of an encoder (42) converting input character string data into compressed character string data, and the decoder (44) converting, into character string data, the compressed character string data output from the encoder. The speech recognition apparatus (2000) inputs an audio frame to the first model (10), inputs, to the second model (20), compressed character string data output from the first model (10), and thereby generates character string data corresponding to the audio frame.