As with any technology things can go wrong. Here we discuss problems we are aware of. Please send email to Ken Kahn <toontalk@gmail.com> if you encounter any problems or have questions.
The guide has been well-tested on Chrome. Some problems have been observed with FireFox and Safari so switch to Chrome (or Edge) if you encounter problems or very slow responses.
The first time the guide is loaded into the browser it should ask for permission to use the microphone or camera. If you need to change the permission then read this help page by Google.
We have occasionally seen where a program that generates speech successfully suddenly stops generating sound. Saving your work, closing the tab, and then reopening the tab seems to fix this.
Training a computer to recognise images runs quickly on modern PCs because it is able to use the GPU that is part of nearly all computers. There is a GPU on a Raspberry Pi but you may find it runs very slowly. On a laptop training takes 100 to 200 milliseconds per image and prediction takes 40 to 60 milliseconds. On an Android phone training ranges from 60 to 1500 milliseconds and prediction takes 1.1 to 1.3 seconds. On a Raspberry Pi training takes 80 to 250 seconds and prediction 100 to 2500 seconds (2 or 3 minutes is typical). At the time of writing this developers are actively working on the Android version of deep learning and not all examples work yet.
Sometimes after many remaking of models or repeated training sessions it begins to predict zero for everything. Saving your work and reloading the web page seems to fix the problem. This chapter has been well-tested on Chrome on PCs and may or may not work on tablets and smart phones.
The Raspberry Pi may think there are multiple audio devices. Right clicking on the speaker volume icon should display alternative devices.
Depending upon the kind of microphone connected (internal or USB) the browser may not know how to access the microphone.
You may need to launch Chromium with an argument describing the alsa-input-device:
chromium-browser --alsa-input-device=...
The Raspberry Pi cannot run Chrome but can run Chromium instead. One of the few ways in which Chromium differs from Chrome is that it has no builtin voices. The easiest workaround is to rely upon the MARY TTS speech blocks instead.