RESEARCH 1 페이지 > Humelo 휴멜로

RESEARCH

Adaptive Prediction for Diversity Promoting Text Generation

2019-11-27
Abstract - Current text generating models are suffering from dull and repeated generation. They are biased to generate frequently used and easily predictable words. This facts mainly contribute to the quality degradation and they hamper applications of text generating models. We propose HumeloGPT to…

READ MORE

A single speech system that can control emotions considering the process of an attention in neural networks and supports two or more speakers in one

2019-11-27
*Patent pending •Introduction Conventional speech synthesis techniques use audio post-processing methods or statistical machine learning with a large amount of data for simulating emotions. Each of these approaches have their own problems. Audio post-processing methods provide flexibility by allowin…

READ MORE

Multi-task Speech Semantic Recognition having Few-shot Behavior

2019-11-27
Because semantic information such as arousal, valence, or abstract emotional label(i.e. happy, sad, angry…) are often manually gathered, speech emotion recognition(SER) research usually suffers from scarcity of emotional label of speech. Our few-shot technologies can be applied to this data scarcity…

READ MORE

Introduction of Humelo’s Lyrics AI

2019-11-27
The "Lyrics AI" learns the lyrics of the song, and creates the lyrics in the way you want without the limit. Below is the advantage of our technology: - The Lyrics AI can create as much lyrics as the user wants for the genre the user desire. The user can also create lyrics with some simple keywords …

READ MORE

Polyphonic Sound Event Detection Using Convolutional Bidirectional Lstm and Synthetic Data-based Transfer Learning (Accepted to ICASSP 2019)

2019-11-27
*Patent Enrolled•Introduction Sound event detection(SED) can be performed to label the datasets derived from these sounds by predicting sound events in the given audio. SED techniques have been applied to audio surveillance, acoustic monitoring for social care, urban sound analysis, multimedia event…

READ MORE

Multi-instrument Composition AI from motif melody

2019-11-27
The composition AI takes the melody as an input and extracts information such as tempo, octave, and chords. And the AI uses it as a motif to play a string of notes. Beyond making new music by arranging songs by real musicians such as XXX and Sleepy before, The goal is to create AI that can compose m…

READ MORE

Phonemic-level duration control using attention alignment for natural neural speech synthesis (Accepted to ICASSP 2019 oral)

2019-11-27
Nowadays, attention-based neural speech synthesis systems such as Tacotron 2 successfully achieve human-level good quality. These recent breakthroughs, however, have some limitations 1. Controllability •Unlike human, attention-based speech synthesis produces the same output for the same input text. …

READ MORE

Harmonic Embedding Methods for Composing/Arranging AI based on Sequence-to-sequence(seq2seq) ANNs.

2019-11-27
* Patent pending Although many generative neural networks are introduced nowadays, neural networks for composing/arranging AI are suffered from “inharmonic generation problem”. Co-founders of Humelo Inc, Jungbae Park and Jaryong Lee developed new embedding methods for composing/arranging seq2seq neu…

READ MORE