Summary
AI avatars have become convincing on screen, but their voices often still sound like they were recorded in a closet. Julia McCoy tests HeyGen Voice, a new voice model built to adapt to video, using her personal avatar, which has produced more than 750 videos she never filmed. The video is sponsored by HeyGen, and she says the test and opinions are her own.
HeyGen Voice clones how a person sounds, then adjusts delivery to match the environment shown on screen. McCoy plays the same line in a studio, on a beach, on a busy city street, at a distance, and on a stage with mic echo, then challenges viewers to place each voice without the picture. She also explains the Lombard effect, named after French ear doctor Etienne Lombard, who noticed in 1909 that people raise their voices in noisy settings.
She notes the model debuted at the top of the Artificial Analysis voice leaderboard, ahead of ElevenLabs V4 and Alibaba’s CosyVoice, while cautioning that leaderboards change. The video also covers an expressiveness boost slider, practical uses for keynotes, walk-and-talk videos, and client-facing content, and the steps to run the same test on your own avatar.
📺 Source: Julia McCoy · Published November 08, 2026
🏷️ Format: Tutorial Demo
