TL;DR
This study investigates how vocal training impacts the acoustic features of amateur male singers and how these changes affect perceived vocal quality. The research reveals that training significantly improves vibrato and the "singing formant," leading to a more pleasing vocal sound, with vibrato being the most influential factor.
What Problem Does This Paper Address?
Amateur singers often seek vocal training to improve their singing quality, but the specific acoustic changes induced by training and their perceptual impact remain unclear. While previous research has identified acoustic features associated with professional singing (like vibrato and the singing formant), the direct link between vocal training, these features, and the subjective perception of "good singing" in amateurs hasn't been thoroughly explored. This paper aims to bridge this gap by analyzing the acoustic transformations in amateur singers after training and quantifying their effect on perceived vocal quality.
Methods
The researchers employed a multi-faceted approach:
- Acoustic Analysis: They recorded amateur male singers before and after a short period of vocal training. Using the STRAIGHT analysis tool, they meticulously analyzed fundamental frequency (F0) contours and spectral envelopes. Specific focus was placed on four key acoustic features unique to singing: vibrato (rate and extent), overshoot (extent and duration), preparation (extent and duration), and the "singing formant" (a spectral peak around 3 kHz).
- Psychoacoustic Experiments: Two experiments were conducted to assess the perceptual impact of these acoustic changes.
- Experiment 1: Used STRAIGHT-morphing to gradually transition between untrained and trained voices, both in terms of F0 contour and spectral envelopes. Participants rated the perceived vocal quality of these morphed stimuli.
- Experiment 2: Employed a speech-to-singing synthesis system to create controlled stimuli. This system allowed researchers to isolate and manipulate individual acoustic features (vibrato, overshoot, preparation, singing formant) by converting them to "professional" characteristics while keeping others "untrained." Participants then compared these synthesized voices.
Datasets
- Singing Voice Recordings: Recordings of three amateur male singers singing two Japanese popular songs before and after nine hours of vocal training. A professional tenor's recordings were also used as a reference.
- Psychoacoustic Stimuli:
- Experiment 1: 66 stimuli per song (11 morphing rates x 3 singers x 2 songs) created by morphing F0 contours and spectral envelopes.
- Experiment 2: 180 stimuli per song (6P2 combinations of 6 synthesized voice types x 3 singers x 2 songs) where individual acoustic features were manipulated.
Key Insights and Ideas
- Vocal Training Works (Acoustically): Vocal training demonstrably alters key acoustic features in amateur singers, particularly improving the steadiness and extent of vibrato and enhancing the singing formant.
- F0 is King (Perceptually): Changes in F0 contour (like vibrato and overshoot) have a more significant impact on perceived vocal quality than changes in spectral characteristics alone.
- Vibrato is the Star: Among the analyzed features, vibrato emerged as the most crucial for perceived vocal quality, followed by the singing formant, then overshoot, and finally preparation.
- Training Mimics Professionalism: The acoustic changes observed in trained amateurs showed a trend towards approximating the characteristics of a professional singer.
Key Results
- Acoustic Improvements:
- Vibrato: Extent increased, and variance decreased (making it steadier). Rate was less affected.
- Overshoot: Extent and duration were reduced, approaching professional levels for some singers.
- Singing Formant: Spectral peaks near 3 kHz increased by an average of 13 dB after training, suggesting the development of a singing formant.
- Perceptual Findings:
- Experiment 1 showed that as morphing rates increased (moving towards the trained voice), perceived vocal quality improved, with F0 morphing having a stronger effect than spectral morphing.
- Experiment 2 revealed a clear hierarchy of perceptual importance for acoustic features: Vibrato > Singing Formant > Overshoot > Preparation. Specifically, converting vibrato to a professional standard yielded a greater perceived quality improvement than doing so for overshoot, preparation, or singing formant individually.
Limitations and Future Work
- Limited Training Duration: The vocal training period was relatively short (9 hours over 3 days). Longer training durations might yield more pronounced or different acoustic changes.
- Small Sample Size: The study involved only three amateur male singers. Future work should include a larger and more diverse group of singers (including females).
- Focus on Specific Features: While the paper covers key features, other aspects of vocal production and perception could be explored.
- Future Directions: The authors suggest further detailed investigation into the relationship between acoustic features and vocal training, with the ultimate goal of developing a high-quality singing voice synthesis system. They also plan to explore these effects in more detail.
Summary updated