Voice Cloning tool from OpenAI
โข OpenAI has built a voice cloning tool called Voice Engine that can generate synthetic speech matching any voice from a 15-second sample. โข Voice Engine powers the voice capabilities in ChatGPT and OpenAI's text-to-speech API, and has been used by Spotify for podcast dubbing. โข The model was trained on a mix of licensed and publicly available speech data, though details are not provided. โข Voice Engine generates speech on-the-fly without building custom models, allowing cheap pricing around $1 per hour of audio. โข It lacks controls to adjust characteristics like pitch and tone, though it aims to mimic the expressiveness of the sample voice. โข The tool could commoditize voice acting work, though OpenAI is exploring actor compensation models. โข Voice cloning carries risks like harassment, fraud, and election interference via deepfakes. โข OpenAI is limiting initial Voice Engine access and use cases while exploring mitigations like watermarking. โข Future plans include security by having users read randomized text to prove consent. โข OpenAI is reluctant to commit to a general release until safety issues from the pilot are understood. https://techcrunch.com/2024/03/29/openai-custom-voice-engine-preview/