This week has been quite busy at work but I have had some time at working on my Chinese as well as an audio side-project.
During the week I found a post on chinese-forums.com where someone described his approach to learning 4000+ Chinese characters. I’m still at around 2000 where I’ve been stuck a while. I thought I could increase my vocabulary by primary reading, but I’ve realised that this maybe isn’t enough by itself.
The method he is using is similar to how I studied when I started learning Chinese back in 2014, but he stuck with it for far longer than me. Basically, you learn a few stories related to the components of the characters (I used Learning Chinese Characters - Tuttle). This guy continued creating stories for over 3000 characters more.
I read it and started Googling about memory techniques for learning Chinese and stumbled upon this post by Alex Mullen, a memory champion, which builds upon this system by Serge Gorodish, an American mathematican.
After reading I started working on my own encoding for the pronunciation of characters. I am still adapting it to famous people I’m more familiar with.
Realtime audio transcription
Since I want to learn more about machine learning for audio I thought I could try my hand on creating a self hosted realtime transcription service from the browser.
I started last weekend and have been working on it for a few mornings during the week.
Facebook recently released a pretrained self-supervised model (Wav2vec 2.0) which I have been playing around with for a bit.
To stream audio directly from the browser to a server I’m using a python implementation of webRTC called aiortc.
I’ve run into some problems with not getting very good audio quality which I hope to figure out the coming week.
Since I’m responsible for a new project I also need to start pitching the project internally. I did a first presentation for a few higher-ups at HSBC this Friday. After the presentation I got a recommendation to take a look at two videos about pitching: