We now have Persistent Circles. You can create a Circle in advance to let people know when they should join you. Or you can leave a recurring Circle in place.
Together version 1.0 is out. It has been a long road and we are very happy.
A new version of Together Test is ready for people to use. It is beautiful and robust and has no bugs that we know of. I just sent out a message to the Togethersphere Users Facebook group inviting people to come in and try it out.
To begin with, this website is new. It is incomplete in both content and formatting, so as they always say "bear with us". It has been mentioned to me that some web browsers bring up scary warnings when people try to look at it. I do intend to convert it to https, which should make those warnings go away, but I don't feel like doing that right now. Meanwhile, be assured that nobody is asking you for any passwords or information of any kind.
For some time now we have been thinking that Together is ready for early release, which we call "Version 0". And day after day, week after week, we discover that there is still "one last thing" that needs fixing. Right now we have several people who live in various parts of the world trying to install and run Together on their Mac or Windows computers. We've had some success and found some more problems to fix. So not ready for full release today.
Guillaume is sitting beside me, migrating all of our Windows libraries from MINGW to MSVC. It is a painful process, but a couple of Windows install failures revealed to us that this pain is necessary.
Guillaume just finished adding the RNNoise neural denoiser to microphone input in Together. A denoiser takes the audio from your microphone and applies a filter to remove background noises. If you are calling from a noisy environment like a coffee shop, then the denoiser has a lot of work to do. There is a trade-off between the amount of background noise removed and the quality of your voice that remains. So if you call from a really noisy coffee shop we might be able to filter out the clinking and banging and general hubbub, but your voice will sound flat and even garbled as a result. A Quebec researcher named Jean-Marc Valin created a new, smarter denoiser that trains artificial intelligence to discriminate between background noise and the human voice. We've been testing it for a year now and found that it does an incredibly good job in a coffee shop. We also found that it does flatten the voice a bit, so we decided to use a more conventional denoiser from WebRTC. The WebRTC denoiser leaves the voice sounding rich and beautiful, but doesn't remove a lot. And now, with this latest addition, you can use WebRTC denoising for a normal call in a quiet room and choose an option to add the aggressive denoising of RNNoise only if you need it.
by Barbara Samson