Audio Intent Recognition Bypassing Text Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current intent recognition methods in human-machine dialogue systems rely on transcribing audio to text, which is prone to errors due to background noise or accent issues, leading to decreased accuracy and efficiency in intent recognition.
Innovation Solution
The method involves acquiring user audio, determining multi-frame audio feature vectors, and inputting these vectors into a preset intent recognition model to obtain output sequences, which include blank and non-blank characters, to directly determine the target intent without text conversion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If audio is transcribed to text before intent recognition, then text-based processing can be performed, but recognition accuracy decreases due to conversion errors from background noise or accent
Solution Approach 1:
The patent extracts and removes the text conversion step from the traditional audio-to-intent pipeline. By directly processing audio features through the intent recognition model, it eliminates the intermediate text transcription that causes information loss and accuracy degradation.
Solution Approach 2:
The patent introduces audio feature vectors as an intermediary representation between raw audio and intent recognition. These feature vectors capture essential audio characteristics without requiring text conversion, serving as a effective mediator that preserves information while enabling model processing.
2Productivity
If audio is transcribed to text before intent recognition, then standard text processing can be used, but processing time increases due to the text conversion step
Solution Approach 1:
The patent removes the time-consuming text transcription step from the processing pipeline. By directly feeding audio feature vectors into the intent recognition model, it eliminates the conversion bottleneck and achieves faster processing.
Solution Approach 2:
The patent segments the audio signal into multiple frames and extracts features from each frame independently. This segmentation approach enables parallel processing of audio features, significantly reducing the overall processing time compared to sequential text conversion.
3Measurement precision
If text conversion is performed for intent recognition, then text-based analysis can be conducted, but conversion errors from noise or accent reduce recognition quality
Solution Approach 1:
The patent transforms the harmful effect of background noise and accent by processing audio features directly rather than converting to text. The audio feature extraction process inherently handles noise and variation better than text conversion, turning the challenge of noisy audio into an advantage by preserving the raw audio characteristics.
Solution Approach 2:
Audio feature vectors serve as an intermediary that is more robust to noise and accent variations than text representations. These features capture the essential characteristics of speech while being less sensitive to adverse conditions, improving recognition precision.
Data Source
AI summary
An intent recognition method, an intent recognition apparatus, a storage medium and a computer device are provided, aiming to improve the accuracy and the efficiency of the intent recognition, which relates to the technical field of information. The method includes: acquiring user's audio to be recognized; determining multi-frame audio feature vectors corresponding to the audio to be recognized; inputting the multi-frame audio feature vectors into a preset intent recognition model to obtain multiple output sequences corresponding to the multi-frame audio feature vectors, wherein the output sequences comprise blank characters and non-blank characters; and determining a target intent corresponding to the audio to be recognized based on the multiple output sequences. The present disclosure is suitable for recognizing the intent of the user.