Audio Similarity Search With Reference Attributes and Time Visualization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio search methods are inconvenient for users to handle fine numerical values of audio attributes, lack flexibility in setting search conditions, and provide insufficient visualization, leading to deteriorated user experience and inaccurate results.
Innovation Solution
An information processing apparatus and method that uses a deep learned DNN model to estimate similarity degrees between input audio and search conditions based on user-designated audio attributes, allowing flexible condition setting and intuitive visualization of results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If an audio processing apparatus detects audio sections based on search keywords, then audio search can be performed, but the search lacks flexibility when users need to specify other conditions beyond keywords and cannot handle input mistakes effectively
Solution Approach 1:
The system creates a reference audio example that users can select instead of manually specifying complex search parameters. By copying the characteristics from the reference audio, the system automatically determines appropriate search conditions, making the operation easier while maintaining flexibility.
Solution Approach 2:
The reference audio serves as an intermediary between the user's intent and the search algorithm. Instead of directly specifying technical parameters, users provide a reference audio that mediates the translation into search conditions, resolving the contradiction between flexibility and ease of use.
2Measurement precision
If an audio search method focuses on prosody features to find similar audio sections, then comprehensive audio matching is achieved, but the method lacks flexibility to search for audio matching only specific attributes like reverberation time or speaker
Solution Approach 1:
The system segments the audio comparison into multiple independent attribute dimensions (prosody, reverberation, speaker characteristics). Users can select which dimensions to match, allowing precise control over which attributes are considered while maintaining the ability to perform comprehensive matching when desired.
Solution Approach 2:
The matching criteria are made dynamic and adjustable. The system can adapt between comprehensive prosody-based matching and selective attribute-based matching depending on user needs, providing flexibility in determining the degree and type of similarity required.
3Measurement precision
If users handle fine numerical values of audio attributes directly, then precise search conditions can be set, but user experience deteriorates due to the inconvenience of handling numerical values
Solution Approach 1:
Instead of requiring users to input numerical values directly, the system copies the desired characteristics from a reference audio example. This eliminates the need for users to handle numerical values while maintaining precise search conditions based on the reference audio's actual properties.
4Productivity
If audio search results are provided without visualization, then processing speed is maintained, but user understanding of similarity degrees and result selection becomes difficult
Solution Approach 1:
The system uses visual representation with color coding to indicate similarity degrees. Different colors or visual intensities represent different levels of match quality, allowing users to quickly understand and compare results without losing information about the degree of similarity.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
Provided is an information processing apparatus that searches for an audio section having a specific attribute from an audio material. The information processing apparatus includes a search condition setting unit that sets a search condition based on a combination of a reference audio and a type of an audio attribute, and a similarity degree estimation unit that estimates a similarity degree between an input audio and the search condition. The similarity degree estimation unit estimates a similarity degree between the input audio and the reference audio on the basis of a type of an audio attribute by using a learned model. The information processing apparatus further includes a visualization control unit that performs control so as to display a transition of the similarity degree estimated by the similarity degree estimation unit along a time axis of the input audio.