Partial-Audio Recognition Using BM25-BERT Candidate Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users often fail to identify songs they hear due to lack of knowledge about song names or singers, hindering effective music retrieval and reducing user experience.
Innovation Solution
An audio recognition method involving a trained detection model that processes segment information from a query content, selects candidate audios from a library, and uses a BM25 algorithm and BERT model to identify target segment information and audios, enhancing retrieval efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If users retrieve music based on song title or singer, then retrieval can be performed, but users cannot identify songs they hear without knowing song names or singers
Solution Approach 1:
The patent replaces manual text-based search operations with automated audio recognition technology. The system uses audio fingerprinting and machine learning models to automatically identify songs from audio segments, substituting the mechanical process of typing song titles with an automated acoustic analysis system that works behind the scenes.
Solution Approach 2:
The system enables self-service audio recognition where the detection model automatically processes audio segments and identifies target songs without requiring user intervention. The model autonomously compares audio fingerprints, ranks candidate songs, and returns results, allowing the system to serve itself in the recognition process.
2Measurement precision
If a trained detection model is used to process audio segments, then recognition accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
The system performs preliminary action by pre-computing audio fingerprints for all candidate songs and storing them in advance. When a recognition query arrives, the system only needs to compare the query's fingerprint against pre-computed references, significantly reducing real-time processing requirements while maintaining high accuracy through the trained detection model.
3Productivity
If candidate audios are selected from a preset library, then retrieval efficiency is improved, but the complexity of managing the library increases
Solution Approach 1:
The system uses copying by creating audio fingerprint representations of songs rather than storing and managing complete audio files. These compact fingerprint copies enable efficient comparison and retrieval without requiring the system to handle full audio data, reducing library management complexity while maintaining retrieval efficiency.
Data Source
AI summary
A method includes obtaining a query content. The query content includes segment information representing a to-be-recognized audio. The method further includes selecting the preset quantity of candidate audios corresponding to the query content from a preset library. Each candidate audio includes a candidate audio segment matched with the segment information. The method further includes inputting the candidate audio segment into a trained detection model so as to obtain target segment information including the segment information and a target audio where the target segment information is located.