"Giving the Gift of Voice to the Hearing Impaired" KT Puts Heart into AI

[Asia Economy Reporter Koo Chae-eun] KT announced on the 26th that it delivered a one-of-a-kind voice created through the ‘Finding Voice’ project to the participants. The participants' voices were completed through the efforts of their families and KT's artificial intelligence technology.

Finding Voice is a project that creates voices for deaf individuals who have lost their hearing or have lost their voice due to accidents or illnesses. It started from the concern to alleviate daily inconveniences with technology that helps improve life. KT selected 20 participants in April and began voice implementation.

KT possesses the nation's top-level personalized text-to-speech technology (P-TTS). Personalized text-to-speech technology is a technology that creates a person's voice through deep learning-based training. In this Finding Voice project, KT implemented voices without any personal voice training data for the first time in Korea. Existing speech synthesis technology required at least one sentence recorded in the person's own voice. However, KT created voices using family voice data for deaf participants who found it difficult to produce their own voice.

The participants' voices reflected the vocal tone, intonation, and speech style unique to each participant based on voice data from same-gender family members. KT analyzed individual characteristics such as gender, age, and oral structure with an AI engine to create distinctive voices for each participant. Each same-gender family member recorded 1,000 sentences for the voice implementation, taking an average of six hours per person.

KT conducted voice modeling using family voice data and oral structure data as two axes. When siblings with similar oral structures recorded, the error in the implemented voice values was small, making voice implementation relatively smooth. However, when parents recorded the voice, the discrepancy between the new voice and oral structure was large, often requiring new modeling. Additionally, intonation differences due to age also needed correction.

KT developed a dedicated mobile application called ‘Maeum Talk’ (hereafter Maeum Talk) so participants can always communicate with their implemented voices. Maeum Talk is a service available only to Finding Voice participants and their families and acquaintances. Maeum Talk converts text input by deaf users in the app into each participant's voice by transmitting it to KT's GPU cloud platform. During this process, tens of millions of calculations are performed on GPUs to generate the speech. The actual computation time is about one second, allowing users to have real-time conversations without perceivable delay. Frequently used sentences can be saved and played instantly when needed to convey voice to others. It also assists communication when deaf and hearing individuals are in the same space.

One of Maeum Talk’s features, ‘My Voice Voice/Video Call,’ allows deaf users to communicate by typing messages while the other party talks as in a regular voice call. During a voice call, users can switch to a video call without disconnecting, enabling communication using both sign language and voice simultaneously. KT plans to support the dedicated app for the next two years and will continuously update the app by checking user inconveniences.

Ms. Song Jae-hwa, mother of Kim So-hee who appeared in KT’s March corporate advertisement ‘My Name is Kim So-hee’ from the ‘Filling the Heart’ campaign, said, “Because my eyesight is poor, I couldn’t see So-hee’s messages well, so when So-hee went out and contacted me, my granddaughter had to act as an intermediary messenger,” adding, “Using the app, So-hee and I can talk directly, which is convenient, and hearing my daughter’s voice is very nice.”

This content was produced with the assistance of AI translation services.