The firm detailed the brand new AI instrument in a blog post. Hume said that it’s attempting to resolve the issue of enterprises discovering the proper AI voice to match their model id. With this function, customers can customise totally different elements of the notion of voice and permit builders to create a extra assertive, relaxed, or buoyant voice for AI-based functions.
Hume’s Voice Control is at present accessible in beta, however it may be accessed by anybody registered on the platform. Gadgets 360 employees members had been capable of entry the instrument and check the function. There are 10 totally different dimensions builders can alter together with gender, assertiveness, buoyancy, confidence, enthusiasm, nasality, relaxedness, smoothness, tepidity, and tightness.
Instead of including a prompt-based customisation, the corporate has added a slider that goes from -100 to +100 for every of the metrics. The firm said that this method was taken to get rid of the vagueness related to the textual description of a voice and to supply granular management over the languages.
In our testing, we discovered altering any of the ten dimensions makes an audible distinction to the AI voice and the instrument was capable of disentangle the totally different dimensions accurately. The AI agency claimed that this was achieved by creating a brand new “unsupervised method” which preserves most traits of every base voice when particular parameters are assorted. Notably, Hume didn’t element the supply of the procured information.
Notably, after creating an AI voice, builders must deploy it to the appliance by configuring its Empathic Voice Interface (EVI) AI mannequin. While the corporate didn’t specify, the EVI-2 mannequin was seemingly used for this experimental function.
In the long run, Hume plans to broaden the vary of base voices, introduce further interpretable dimensions, improve the preservation of voice traits beneath excessive modifications, and develop superior instruments to analyse and visualise voice traits.
