Humelo Elevates AI Voice Quality to Studio Level

A new technology has emerged that elevates the audio quality of AI text-to-speech (TTS) from standard call quality to studio-level sound, comparable to high-quality streaming audio.

Voice AI startup Humelo announced on August 14 that it has upgraded its TTS service audio quality from the existing 24kHz to 48kHz using its independently developed ultra-fast upsampling technology.

The audio quality of digital sound is determined by the sampling rate (Hz), which indicates how finely the sound signal is divided per second. Typically, the 16kHz audio quality used in phone calls or standard AI chatbots allows voices to be understood, but background noise is present, and the sound tends to feel mechanical. In contrast, 48kHz is the standard specification used in most media content production today, including music streaming, movies, broadcasting, and games, delivering clear and richly textured sound.

Until now, the industry has been reluctant to implement high-quality TTS at 48kHz, as it is difficult to secure large amounts of original high-quality audio data, and the enormous computational power and infrastructure costs required for high-quality processing would lead to higher service prices. To address this issue, Humelo independently developed a "voice super-resolution upsampling" technology that restores low-quality audio data to ultra-high-resolution, high-quality sound. With this newly unveiled technology, even very low-quality audio at 8kHz can be enhanced to studio-level 48kHz quality.

Kwon Yongseok, CEO of Humelo, said, "When TTS technology was less advanced, demand was low, and as the technology improved, the cost issue slowed the adoption of high-quality sound. Humelo's independently developed ultra-fast, high-quality voice AI technology will open up new experiences and possibilities at reasonable costs for all creators and companies who have hesitated to adopt high-quality TTS due to cost concerns."