What It Does
CVoice starts from cues such as fundamental frequency, loudness, spectral shape, rhythm, and phonation stability, and summarizes them into eight directions: pitch tendency, brightness, softness, weight, liveliness, stability, warmth, and clarity. It then compares these directions with a set of human-designed voice prototypes to generate a primary timbre, combined labels, and entertainment-oriented descriptions.
The analysis also routes sustained tones, whistles, humming, airflow noise, and other special sounds separately, so clips that are clearly not natural speech are not forced into ordinary voice rankings.
How to Make a More Useful Recording
- Use 10–30 seconds of natural speech at your normal pace; there is no need to deliberately lower or raise your voice.
- Keep the microphone about a palm’s width away, avoid fans, music, and strong reverberation, and prevent waveform clipping.
- When comparing two results, use the same device, the same position, and similar text whenever possible.
How to Interpret the Results
Population shares come from anonymous classification counts. They only indicate how common a result is within the current sample and do not imply quality, value, or that rarer voices are better.
Recordings & Data
Original recordings are deleted after analysis completes or fails. Tasks and results are retained for up to about 24 hours so they can be restored after a page refresh. Anonymous aggregates record only classifications and counts, not recordings or personal identity.