With the advancement of deep learning technology,
more complex and acccurate speech analysis and processing have become possible.
Speaker verification
Speaker verification is the process of determining if the input voice matches the specific user’s voice, calculating the distance between voice footprints. Similar to how facial recognition functions based on the arrangement of facial muscles, speaker verification characterizes the voice by recognizing the timbre’s formant placement, which is influenced by the tension state of the muscles linked to the vocal tract.
One-shot voice conversion
We use signal processing techniques and various deep learning techniques to extract linguistic (content) information of speech and voice feature information of the speaker. The extracted information is synthesized and used to reconstruct high quality and high similarity speech.