Dear backers,
We hope that you and your family are safe.
In Shenzhen, life is almost back to normal: all the shops are opened and the whole team is at the office. And while everyone still wear masks, people are getting their life back.
We haven’t given much news but we’ve been working every day to prepare mass production and finish developing the teaching software.
Setting up the mass production assembly process
The production of your speakers doesn’t happen at one place. Components and the assembly of those components are scattered all over the place. For example, to assemble Lily’s plastic parts, fabric and loudspeaker, we need to go through 4 different factories as you can see below:
Our team needs to go on-site to many of those factories to prepare assembly and teach them how we want the quality control to be done.
Developing the mass production testing software
We did PCBA tests and performance tests (drop/temperature tests) during the development phase of the product to make sure that the product was working. Now we’re developing the testing software for mass production to control the quality of the product.
For example, to test the sound quality, we put Lily in this acoustic box and we run a testing software (from inside the box) to test the frequency response, impedance, signal-to-noise ratio and distortion of Lily’s sound system.
Aside from sound quality, we’re developing mass production testing software for the touch gestures, LEDs animations, Bluetooth, Wi-Fi pairing, etc…
Problems with our Wake Word data collection
In a previous update, we talked about our wake word data collection and how we contracted with a voice data provider that would record the wake word at scale.
Unfortunately this collaboration has been catastrophic. The voice data we received had missing files and inconsistencies (2 different persons with the same voice). We asked for more than 60% native English speakers (US, England, Canada, etc..) but the provider gave us recordings from Liberia. And while Liberia’s official language is English, Liberian English accent doesn’t reflect our user base.
We had to cancel our collaboration, fight to get refunded and contract with a new voice data provider. We were paying the 1st provider $10/person and now we are paying the new data provider $56/person (each person records 9 wake words and 45min of speech data). We’ve also developed our internal voice data collection app (that our voice data provider will use) to ensure that the voice collection is done right.
In parallel, we’re also developing keyword spotting models that rely on less data (while maintaining > 90% accuracy).
Working on Language Understanding (Lily’s brain)
After transcribing your voice into text (Speech Recognition), Lily needs to analyze the transcribed text to understand what you want. This part of Lily’s AI is called Natural Language Understanding and we’re currently working on it. Here’s a picture that shows how it works:
Note that this is a very simplified version of how Lily processes human language, used for explanation purpose only.
Lily’s app and teaching system
We’re making lot of progress on Lily’s app (used to control the speaker, launch the lessons, see the training program, see the characters & pinyin, etc…) but we won’t be sharing any more screens of our app since we’ve already been copied and don’t want to be copied anymore. We’re also currently patenting our voice teaching system.
Lily’s app, the Chinese curriculum and the voice exercises are a major part of what we’re working on.
Timeline and shipping
The hardware should be ready to go in mass production by end of June with a pre-production batch in the middle of June.
So we’re still on track to start shipping at the end of June. We’ll probably start with 100 units that we’ll drop ship and we’ll send you the tracking numbers.
The status of our company
Due to the global economic situation, we’re also under economic pressure, it’s much harder to get financing. The whole team willingly accepted to delay 2019’s end-year bonuses and we’re very careful on burn rate. We didn’t renew our community manager’s internship in January (hence the slower responses in messages) but we hired 2 more AI/software engineers. Our team grew a little bit, we’re approx 20 now (including a few US-based engineers). We have the finance, the team and the passion to finish developing Lily and ship it to you.
Thanks for your continuous support.
Jie and the Maybe team
