Revcom Research’s paper on “Emotional annotation of speech using large-scale language models” was selected for the international conference on audio signal processing “ICASSP2024”
*View in browser* *Revcom*
Press release: March 28, 2024
**
Revcom Research’s paper on “Emotional annotation of speech using large-scale language models” was selected for the international conference on audio signal processing “ICASSP2024”
* RevComm, the research and development organization of RevComm Co., Ltd. (Headquarters: Shibuya-ku, Tokyo, Representative Director: Takeshi Aida)
Research (Revcom Research, RCR) will present a paper on “Emotional Annotation of Speech Using Large-Scale Language Models” at the world’s largest conference in speech and acoustic signal processing, which will be held from April 14th to 19th, 2024. International academic conference “ICASSP”
2024” (Seoul, South Korea). *
*What is ICASSP*
“ICASSP” (International Conference on Acoustics, Speech, and Signal) IEEE Signal Processing) is a signal processing society with the longest history in the Institute of Electrical and Electronics Engineers of America.
This is an international academic conference sponsored by “Society”, and this year marks the 49th time it has been held.
*Paper content*
As a result of RCR’s research, Senior Research Engineer Jennifer Santoso, Kenkichi Ishizuka, and research director Taiichi Hashimoto published a paper titled “Large
Language Model-Based Emotional Speech Annotation Using Context and Acoustic Feature for Speech Emotion Recognition” was submitted to “ICASSP2024” and was accepted.
Conventionally, in order to add emotional information to voice, a large amount of cost is required to manually listen to the voice, identify the emotion, and add it, making it extremely difficult to create large-scale voice data with emotional information. It was difficult.
In this research, we use a large language model (Large Language Model) to estimate emotions based on speech transcription and phonetic features.
We are proposing a method for automatically granting bonuses using LLM). Experiments in this study showed that large-scale language models can estimate emotions with almost the same accuracy as humans. The results of this research are expected to facilitate the creation of large-scale voice emotion data and enable the development of more accurate voice emotion recognizers.
*Recently accepted paper*
Large Language Model-Based Emotional Speech Annotation Using Context and Acoustic Feature for Speech Emotion Recognition
https://ieeexplore.ieee.org/document/10448316
RCR aims to bring innovation to the AI technology field and enrich communication. To this end, we will continue to promote research and development in the areas of speech, language, and images, and actively work to make academic contributions both domestically and
internationally, and to deepen AI technology for products and services.
* What is “RevComm Research (RCR)” *
RCR is an organization that researches and develops new forms of communication with the aim of creating a society where people can understand each other better. When people communicate with others, friction and unequal situations often occur, leading to situations in which mutual understanding and trust are lost. We will develop technology to eliminate such friction and create an environment that allows for more flexible and appropriate communication without misunderstandings.
“RCR” also includes “Research for Communication”.
It also includes the meaning of “Revolution”. Based on our corporate philosophy of “reinventing communication and creating a society where people care about others,” we work to solve communication issues through research and development of voice technology and AI.
RCR site: https://www.revcomm.co.jp/rcr/
Past activity results: https://www.revcomm.co.jp/rcr/information/ *RevComm Co., Ltd. Company Profile*
Based on the philosophy of “reinventing communication and creating a society where people care about others,” we are a company that uses voice technology and AI to solve communication issues.
The voice analysis AI phone “MiiTel” is used mainly in the inside sales market for visualizing talk, self-coaching, and building telework systems in sales and call center operations. In addition, the AI-equipped online meeting analysis tool “MiiTel”
“Meetings” allows you to analyze and review online meetings with multiple people, dramatically increasing your sales closing rate. “MiiTel” visualizes offline (face-to-face) business negotiations RecPod (α version)” has also been launched, making it possible to convert conversations into big data wherever communication occurs.
Forbes JAPAN “Japanese Entrepreneur Ranking”
In addition to being selected for Weekly Toyo Keizai’s “Awesome Venture 100,” in April 2023, it was the only Asian company to be included in the U.S. “Forbes AI 50”.
In May 2023, we were awarded first place in the Deloitte Tohmatsu Group’s Technology Fast 50 2022 Japan.
・Company name: RevComm Co., Ltd.
・Location: 7th floor, Hulic Shibuya 1-chome Building, 1-3-9 Shibuya, Shibuya-ku, Tokyo 150-0002
・Representative: Takeshi Aida
・Business details: Development of AI x voice software database ・Company website: https://www.revcomm.co.jp/
*Company names and product/product/service names (including logo marks, etc.) listed are trademarks of each company or registered trademarks of their respective rights holders.
*About details about this release*
https://prtimes.jp/main/html/rd/p/000000228.000037840.html
*Download press release materials*
https://prtimes.jp/im/action.php?run=html&page=releaseimage&company_id=37840&release_id=228