Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Voice Calls: Secure, Crystal-Clear, AI-Powered

telegram.org

Post reply on HN
Print view View original

211–220 of 299 posts

Re: Voice Calls: Secure, Crystal-Clear, AI-Powered

#211

Post by Florin_Andrei »

torrent-of-ions wrote: post #121

I will definitely try this because I have trouble with all other voice chat services including Skype, WhatsApp and Hangouts. Often one of those three will work but not always. But the most annoying problem with any internet voice chat is not so much the quality but the latency. Landline phones have noticeably lower voice quality but one can still enjoy a conversation perfectly well. High latency, on the other hand, a…

> But the most annoying problem with any internet voice chat is not so much the quality but the latency. Landline phones have noticeably lower voice quality but one can still enjoy a conversation perfectly well. High latency, on the other hand, absolutely ruins the experience.

Maybe this is because traditional communications were built by people working exclusively in that field - and they eventually became true experts, with a deep understanding of the real issues.

Nowadays Internet chat services are cobbled together by happy-go-lucky engineers who, prior to this, maybe did an ads firehose on Twitter, and after this project will perhaps move on to work on Facebook games or whatever. Smart people to be sure, but their understanding of the field is necessarily superficial.

The fast rate of change in this industry explains many of the things that are ailing it currently.

Re: Voice Calls: Secure, Crystal-Clear, AI-Powered

#212

Post by idanoeman »

Johnny_Brahms wrote: post #109
lucb1e wrote: post #101

Earlier quoted context omitted.

The encryption they claim to use. Not saying they don't, just saying it's only "sort of" superior on encryption, given this caveat.

The same goes for Telegram. The android client hasn't seen any updates to it's github repo since sep 2016. And, it is quite easy to verify that WhatsApp's encryption is doing what we think. A friend of mine managed to reverse engineer their protocol in 2012 in less than 24 hours, by himself. And there is quite a big chunk of the computer security market that would disagree with the claim that something has to be open…

Couldn't WhatsApp faithfully implement the protocol, but also keylog and phone home occasionally? How would you know?

Re: Voice Calls: Secure, Crystal-Clear, AI-Powered

#213

Post by Animats »

The emoji out of band authentication is cool, but probably annoying. You have to read those emoji by voice to the other end, so they can check them. That could be a pain if the emoji are chosen randomly from the 2600 available emoji.

The idea comes from the STU-3 secure phone, where there was a 2-digit number display to be read back by voice. It's one of the ways to detect a man-in-the-middle attack. If there's a MITM, the crypto bits sent and received are different, because the MITM is re-encrypting, and this is detectable if you have some out of band channel for comparing them. A MITM would thus have to be able to fake the voice of the other party.

With techniques like this, you can make an MITM work arbitrarily hard to maintain the illusion that it's the other party. I've proposed some ways to do this for web pages.

Re: Voice Calls: Secure, Crystal-Clear, AI-Powered

#214

Post by DaiPlusPlus »

Animats wrote: post #210
torrent-of-ions wrote: post #121

I will definitely try this because I have trouble with all other voice chat services including Skype, WhatsApp and Hangouts. Often one of those three will work but not always. But the most annoying problem with any internet voice chat is not so much the quality but the latency. Landline phones have noticeably lower voice quality but one can still enjoy a conversation perfectly well. High latency, on the other hand, a…

The best phone voice quality is still ISDN. Digital all the way to the handset, and rigidly clocked with fixed delay. Switzerland has ISDN phones, and the voice quality is impressive.

ISDN was 64kbps pet channel though - and to my knowledge did not use a wideband voice codec by default (or at least didn't mandate all ISDN receivers support wideband audio).

Lots of people who had ISDN kit plugged their old POTS phones into their TAs anyway.

Re: Voice Calls: Secure, Crystal-Clear, AI-Powered

#215

Post by dmix »

Florin_Andrei wrote: post #211
torrent-of-ions wrote: post #121

I will definitely try this because I have trouble with all other voice chat services including Skype, WhatsApp and Hangouts. Often one of those three will work but not always. But the most annoying problem with any internet voice chat is not so much the quality but the latency. Landline phones have noticeably lower voice quality but one can still enjoy a conversation perfectly well. High latency, on the other hand, a…

> But the most annoying problem with any internet voice chat is not so much the quality but the latency. Landline phones have noticeably lower voice quality but one can still enjoy a conversation perfectly well. High latency, on the other hand, absolutely ruins the experience. Maybe this is because traditional communications were built by people working exclusively in that field - and they eventually became true expe…

> Nowadays Internet chat services are cobbled together by happy-go-lucky engineers who, prior to this, maybe did an ads firehose on Twitter,

Skype has been around for 13 years. It isn't a young startup full of ex-node.js programmers.

WhatsApp also has some excellent infrastructure built on top of Erlang which has decades of experience and refinement of heavy-duty telecom usage baked into the framework.

If anything the experience of previous development trickles down and younger engineers benefit from that knowledge. But I'm not sure I buy the idea that you can only build great systems if you've only been doing that one category for your whole career.

The rate of jobs changes faster and in front-end development the pace of toolset changes pretty fast. But even in front-end development it's still the same languages. usually just the tools are the trendy part. And if it's switching languages every 5-7 yrs that knowledge of languages is transferable. I've only benefited as a programmer from exposure to many other languages (and projects).

Regardless, the hardcore backend guys you get to build this type of infrastructure isn't the same hipster front-end devs that Twitter hired in the early years to build out a consumer product. Consumer is very different than building out telecom infrastructure in Erlang/C++.

Re: Voice Calls: Secure, Crystal-Clear, AI-Powered

#216

Post by backpropaganda »

hannob wrote: post #184

" Each time you make a Voice Call on Telegram, a neural network learns from your and your device‘s feedback (naturally, it doesn’t have access to the contents of the conversation, it has only technical information such as network speed, ping times, packet loss percentage, etc.). The machine optimizes dozens of parameters based on this input, improving the quality of future calls on the given device and network." Is i…

The 'neural network' is probably an SVM or logistic regression. There's nothing special that a neural network can learn with high-level features like network speed, ping times, etc.

In fact, it's probably just a TCP-like algorithm, i.e. decrease rate of packet sending if not getting ACKs back. One day, Intel might come out and say their microprocessors are AI-powered, since it has branch prediction.

Re: Voice Calls: Secure, Crystal-Clear, AI-Powered

#217

Post by l1ambda »

torrent-of-ions wrote: post #121

I will definitely try this because I have trouble with all other voice chat services including Skype, WhatsApp and Hangouts. Often one of those three will work but not always. But the most annoying problem with any internet voice chat is not so much the quality but the latency. Landline phones have noticeably lower voice quality but one can still enjoy a conversation perfectly well. High latency, on the other hand, a…

The Opus codec ( http://opus-codec.org/comparison/ ) can enable very low latency audio. 26.5 ms algorithmic delay (the time it takes to go through the algorithm) by default, but can be set to a very impressive 5ms. Digital audio can't achieve the near no-latency analog POTS has, but it can get pretty darn close e.g. if Opus is used in the very low latency mode. I believe Mumble and maybe TeamSpeak let you choose the Opus latency. I know some other apps like WhatsApp, Wire, and Signal use Opus as well. I don't know what settings they use though.

Another problem is cell phones add ~150ms latency (pre VoLTE) and people that route calls through Google Voice adds an additional 150ms+ latency on top. If you ever talk to someone on a cell phone on google voice, it sounds like they're calling from the moon.

150ms is about the human limit before turn-taking and talking over each other becomes an annoyance. That's why cell-to-landline conversations (150ms one way) were tolerable, but cell-to-cell (300ms one way) conversations are abysmal. And that's all in the same area code. Once you add long distance, it gets worse.

Re: Voice Calls: Secure, Crystal-Clear, AI-Powered

#218

Post by CharlesW »

jdfreefly wrote: post #196
torrent-of-ions wrote: post #121

I will definitely try this because I have trouble with all other voice chat services including Skype, WhatsApp and Hangouts. Often one of those three will work but not always. But the most annoying problem with any internet voice chat is not so much the quality but the latency. Landline phones have noticeably lower voice quality but one can still enjoy a conversation perfectly well. High latency, on the other hand, a…

Have you tried Amazon Chime? We are ridiculously aware of the latency issue in high quality voip. We do not at the moment support user generated keys the way Telegram does but we do use TLS/DTLS encryption end to end for our VOIP streams.

> We are ridiculously aware of the latency issue in high quality voip. We do not at the moment support user generated keys the way Telegram does but we do use TLS/DTLS encryption end to end for our VOIP streams.

Any plans to make this available to developers as a service via AWS?

Re: Voice Calls: Secure, Crystal-Clear, AI-Powered

#219

Post by NoGravitas »

moomin wrote: post #155
torrent-of-ions wrote: post #140

Earlier quoted context omitted.

I don't remember dialup latency. I only remember it having low bandwidth. I'm talking about ADSL latency in the 90s.

Latency for a 56k modem was 180ms. Latency for a voice call on the same line was _much_ smaller. AFAICT, latency from my games console television to my TV is much higher...

300ms was not unusual on a 56k modem, if my memory serves.

Re: Voice Calls: Secure, Crystal-Clear, AI-Powered

#220

Post by pbhjpbhj »

fenwick67 wrote: post #193
moe wrote: post #150

Earlier quoted context omitted.

But what sort of parameters are adjusted? This is HN. A link to an example would be appreciated. Edit: To clarify, I work with audio codecs too, and can't really think of parameters (other than the compression level?) that would make much sense to adjust on the fly. If "AI" is used for more than just a buzzword here, then I imagine the answer must be quite interesting.

They probably adjust the incoming / outgoing buffer sizes (and therefore the audio delay, since it's live) to account for packet loss. They might also prioritize traffic depending on how full your buffers are. I can only assume Youtube and Netflix do similar parameter tweaks to optimize their video delivery based on the connection (totally filling the buffer to a max size all the time would waste bandwidth, but if th…

Right, looks like we're up to maybe 6 parameters. The claim was dozens which i take to mean at least 24, possibly 36 as the lower bound.
Post reply on HN