murchinroom runs a number of projects prefixed with "mu".
The original idea behind these "mu" projects was to develop fun apps related to music.
Now we officially explain "mu" as an abbreviation for "murchinroom universe" "machine & you", which signifies that these projects are exploring the intersection of machine learning (or AI) and human creativity (and emotion).
Let's just forget When I was 12 or something, I programmed a chatbot in the Beginners' All-purpose Symbolic Instruction Code, as my very first code project (excluding hello-worlds and exercises). It was a simple rule-based chatbot, but it is fun and unforgettable.
The autumn night when we realized Now, we are excited to witness the rise of large language models (LLMs) in the past few years. They perform well not only in traditional natural language processing tasks such as text classification and machine translation but also in more creative tasks such as text generation and even chatting with humans.
Not to be overly dramatic As it is getting closer to chat like a human, we are trying to push the boundary further to make it look & sound like a human.
So, let's just pretend Combining the concept of VTuber and LLM, we are building muvtuber: a VTuber chats with you, tells you stories -- driven by machine learning technologies including large language models. No human in the loop.
muvtuber is a integration of text generation, emotion analysis, speech synthesis and Live2D model driver.
Technically, we are trying to fully containerize the system, so that it can be easily deployed on any cloud platform or even become a SaaS product.
Trapped in desert, barely realized While the music recommendation systems (MRS) are widely used today, there is still room for improvement. Existing algorithms perform well for personalized music recommendations, but most of those methods lack humanization, seeing that they are difficult to detect the emotions of users and recommend songs in the mood for the spot in real-time.
No worry, it's a dream at midnight In the murecom project, we designed an emotion-based music recommendation system with the goal of humanization. Through technologies including natural language processing and computer vision, our system analyzes the user's current mood as well as the emotions contained in the music tracks. Recommending related songs in the user's situation, our system generates playlists to accompany the users. Compared with the traditional recommendation algorithm, our system can better meet the emotional needs of users for music.
But dreams and reality have some conformity, don't you be quiet At the same time, we also implemented a recommendation-words writing function based on natural text generation algorithms, allowing the deep neural network to write literary recommendation texts for the playlist output by the MRS just in time. The generated texts focus on the topic of analyzed user mood and recommended tracks, leading users to immerse themselves in the playlist better. Such creative works could only be completed manually, in the past. Therefore our machine learning solution is meaningful to further optimize the user experience of the MRS by making it improved human-friendly.
Technically, murecom is built with a set of microservices. This architecture allows us to easily extend the system with new features as well as to reuse the components in other projects. For example, now we are reusing the services including text emotion analysis (emotext) in the mureader and muvtuber project.
It takes thought, and time, and rhyme On July 1, 1979, Sony released the first iconic cassette Walkman TPS-L2. Impressively, it is equipped with two mini headphone jacks, enabling two people to listen at the same time. At the same time, TPS-L2 has a "HotLine" button to activate a small built-in microphone, which partially overwrites the sound in the tape, allowing one user to talk to another user through music.
In this era when the Walkman has almost become history, occasionally for a moment, we will miss the sense of ceremony and emotion it once brought us. This is the original intention of the birth of musharing.
With music and words l've been flying musharing is a software defined replica of the Walkman TPS-L2. It links people you really care about, and shares music with them synchronously.
Everyone in your room can control the music play/pause, etc, and musharing under the hood will minimize the latency and synchronize the music with all devices.
To be sure that you'll know what I'm saying. musharing embeds a chat panel in the app. You can type to chat with your friends, but it's not all the point. If words can't convey it, just say it. Pressing the "HotLine" button, the volume of the music will be turned down, and the microphone will capture your voice and convey it to your partner.
Won't you let me see what spring is like on Jupiter and Mars Even if you are alone, you don't necessarily have to enjoy the loneliness. We integrate a chatbot in musharing. It understands the music and your feeling. Talk to it.