Sunday 02 March 2025
Scientists have made a significant breakthrough in the field of singing voice conversion, allowing them to transform a source singing voice into that of a target singer while preserving the original lyrics, melody, and various vocal techniques.
The researchers developed a new system called SYKI-SVC, which is capable of producing high-fidelity singing voices. The system consists of three key components: a feature extractor, a voice converter, and a post-processor. The feature extractor uses two models, ContentVec and Whisper, to extract speaker-independent linguistic features from the input singing voice. The voice converter then integrates these features with the target singer’s timbre and F0 (pitch) information to generate the converted singing voice.
The post-processor is where SYKI-SVC really shines. It takes the synthesized singing voice and enhances it by supplementing high-frequency components from the source singing voice. This results in a more natural-sounding output that accurately captures the nuances of the original performance.
To evaluate the system, the researchers created a specialized test set featuring various singing techniques performed by professional singers. The test set included songs with different genres, such as classical, pop, and folk, as well as different vocal styles, like falsetto and bel canto. The results showed that SYKI-SVC outperformed other popular voice conversion systems in terms of naturalness, bite reproduction, technique reproduction, and timbre similarity.
The system’s performance was also evaluated by a group of 10 ordinary listeners and 10 professional musicians who were asked to assess the converted singing voices based on their naturalness, articulation, and overall quality. The results confirmed that SYKI-SVC produced high-quality singing voices that accurately captured the characteristics of the target singers.
One of the key advantages of SYKI-SVC is its ability to preserve the unique vocal techniques of the source singer. This is particularly important for professional singers who want to maintain their distinct sound and style when performing with different accompaniments or backing tracks.
The system’s post-processing component also plays a crucial role in enhancing the quality of the converted singing voice. By supplementing high-frequency components from the source singing voice, SYKI-SVC can accurately capture the subtle nuances of the original performance.
In addition to its potential applications in music production and performance, SYKI-SVC could also have implications for speech therapy and rehabilitation. For example, the system could be used to help individuals with vocal disorders or injuries by generating high-quality singing voices that mimic their natural speaking voice.
Cite this article: “Breakthrough in Singing Voice Conversion Preserves Original Characteristics”, The Science Archive, 2025.
Singing, Voice Conversion, Syki-Svc, Feature Extractor, Voice Converter, Post-Processor, Pitch, Timbre, Vocal Techniques, Music Production
