Singer Identification Using Vocal Separation and Voice Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing music identification techniques often fail to accurately determine that two audio samples are the same song, especially when comparing vocal covers with similar voices.
Innovation Solution
A computing system utilizing machine learning models for music source separation and voice embedding to extract vocal components and compute similarity values between embedding vectors, with silence removal and speaker diarization to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional music identification techniques are used, then the process is simple and fast, but the accuracy is low especially for vocal covers with similar voices
Solution Approach 1:
The music sample is segmented into different sources (vocal component and non-vocal component) using music source separation. The vocal component is further segmented into time segments based on silence removal and speaker diarization. This segmentation allows the system to focus on relevant vocal portions for singer identification, improving accuracy while managing complexity through structured processing stages.
Solution Approach 2:
The system extracts the vocal component from the music sample using music source separation, separating it from instrumental and other non-vocal elements. It then extracts embedding vectors from the vocal component for comparison. This extraction of relevant features (vocal parts and embedding vectors) enables accurate singer identification by focusing computational resources on the most discriminative elements.
2Measurement precision
If music source separation and speaker diarization are applied, then the accuracy of identifying same singer is improved, but the processing time and computational resources increase
Solution Approach 1:
Silence removal is performed as a preliminary action to remove silent portions from the vocal component before embedding vector extraction. This preliminary processing reduces the amount of data that needs to be processed in subsequent stages, potentially reducing overall processing time while maintaining accuracy by focusing on relevant speaking/singing segments.
Solution Approach 2:
The system creates embedding vectors as representations (copies) of the vocal components. These embedding vectors capture the essential characteristics of the vocal parts in a compressed format, allowing for efficient comparison and identification without requiring processing of the entire original audio signal, thus reducing computational time and resources.
3Measurement precision
If embedding vectors are extracted and compared, then the identification accuracy is improved, but the computational complexity increases
Solution Approach 1:
The system transforms the audio signal into embedding vectors, which are parameterized representations that capture essential vocal characteristics. By changing the representation from raw audio waves to compressed embedding vectors, the system reduces the dimensionality and computational complexity of comparisons while maintaining or improving identification accuracy through the use of meaningful feature parameters.
Data Source
AI summary
A computing system including one or more processing devices configured to receive a first music sample and a second music sample. At a music source separation machine learning (ML) model, the one or more processing devices extract a first vocal component from the first music sample and a second vocal component from the second music sample. At a voice embedding ML model, the one or more processing devices extract one or more first embedding vectors from the first vocal component and one or more second embedding vectors from the second vocal component. The one or more processing devices compute a similarity value between the first and second embedding vectors and determine whether the similarity value is above a predefined similarity threshold. Based on the determination, the one or more processing devices output an indication of whether the first music sample has a same singer as the second music sample.