G&L CEO Alexander Leschinsky talks AI-powered sign language for live and on-demand streaming
Some of AI’s signature contributions to streaming video to date have involved subtitling, captioning, dubbing, and translation, where AI/ML and generative AI have dramatically reduced both the associated costs and time required to complete these tasks and thus transformed the landscape for localisation and accessibility. But in the absence of sign-language interpretation, live captioning and audio description for the hearing impaired only opens the door part of the way to full inclusivity, and arguably raises compliance concerns under the regulatory frameworks of some EU member states.
Signapse is a UK-based technology company whose SignStream application provides real-time AI translation for British Sign Language (BSL) and American Sign Language (ASL) through the generation of a photorealistic interpreter in a video stream. In March of this year, German streaming system integrator and Streaming Media Innovation Award winner G&L Systemhaus announced that it had partnered with Signapse to bring to market Digital Signer for German Sign Language (DGS), which Signapse described at the time as “marking a significant step forward in our mission to make real-time sign language accessibility available at scale.”
G&L CEO & Co-Founder Alexander Leschinsky says, “We are excited to extend Signapse’s field-tested, high-performance solution to German Sign Language from autumn 2026 and to harness AI in a way that creates meaningful, real-world impact."
AI Digital Signer for DGS will make its world premiere at IBC2026 on 11 September at the shared G&L and Signapse stand, 14.A42 in Hall 14. On 12 September, Leschinsky will present a session on IBC’s Future Tech Stage, “Broadcast Without Barriers: Live AI German Sign Language.”
In advance of AI Digital Signer for DGS’s IBC2026 debut, I had the opportunity to sit down with Leschinsky to discuss the product, the inclusivity, accessibility, and market needs it addresses, and the challenges of making AI-generated sign language interpretation work in a livestreamed context.
An edited transcript of our conversation follows.
Steve Nathans-Kelly: My guest today is Alexander Leschinsky, co-founder and CEO of G&L Systemhaus, a Germany-based systems integrator who's been building streaming infrastructure for public broadcasters, the European Parliament, and enterprise clients for 25 years. We're here to talk about a project that tackles a longstanding accessibility gap, sign language for live and on-demand streaming. Together with the UK-based company called Signapse, G&L has built a broadcast grade solution that automatically translates spoken or written content into photorealistic AI-powered sign language and integrates it directly into existing streaming workflows. Alexander, thanks so much for joining us.
Alexander Leschinsky: Thanks for having me.
G&L has always been a systems integrator at its core, but AI is very much a part of how you work today. Why did you decide that sign language was a problem worth solving, and why have earlier attempts like 3D avatars fallen short?
At G&L, we are a technical service provider, so we are not sign language interpreters. We don't usually touch the actual content of the customers, but in many of the requests that we get from public entities, especially for event streaming, sign language interpretation was part of the deal. So we got a request and that said, "We have to do this live streaming event. These are the number of languages that we want to support, we need subtitles, and we need a sign language interpreter." As we don't have sign language interpreters, we had to go to the market and see who was available. And we learned from that the hard way that there are only very few of these sign language interpreters, and the more sign language is supported in broadcast and public streaming, the fewer of these sign language interpreters of these humans you get access to because there are simply not enough.
That turned for us into an expensive line item in our invoices or in the offers that we made. More often than not, because we didn't hire some of these interpreters ourselves, we didn't get the job because we had no interpreters that weren't available. You could not book them on short notice; you had to book them for very long hours in advance. For us, it was a technical difficulty getting these into our quotes and into our setup. That was how we learned about the problem in the first place. Because I myself have a background in linguistics and in phonetics and in computer science, I tried to build something three years ago where I looked into the problem. Though the main issue should be getting a realistic 3D animation of a human who is signing, the 3D animation and the character animation with tools like Unreal from the gaming world is not the difficult part.
The difficult part is understanding the grammar, the syntax, and the culture behind it. After having tried to build something ourselves, I quickly saw that this was not going to work. We need experts who know exactly what they're doing. We had to stop our own approach three years ago. So we were happy to find Signapse at IBC in 2025. That was when we first talked to them.
In addition to the technical and linguistic challenges, I believe there's also a compliance aspect to this. My understanding is that sign language interpretation is legally required in more and more contexts across Europe. And the interpreters, as you say, are not there at scale. Can you give me a sense of just how big this gap is and what kind of structural problem it creates?
If you look at the markets that we are currently investigating, because we have support for three languages--American Sign Language, British Sign Language, and soon German Sign Language--you can say that there is a lack of sign language interpreters because we would need more than double the amount that they actually have. So we have way too few of them. And there are many situations where you have a legal requirement to support people who have difficulties in interacting with entities like hospitals or lawyers or law enforcement. If you're a deaf person, you have the right to get a sign language interpreter, but they're difficult to get hold of because there are too few of them.
That is only for the day-to-day requirements that deaf people have. But if you now then see that regulation is forcing many of the broadcasters to increase the amount of content that they provide sign language for, this takes away the scarce resource from the market and blocks the human sign language interpreters that are desperately needed for the interactive situations in law, in hospitals, and with other entities that deaf people have to interact with. If you follow the regulation, it's a good thing because more content is being signed, but at the same time, if you do it with human sign language interpreters, it takes away these few people from the actual bidirectional conversations where deaf people need them desperately. That's the problem that we see.
It does create significant challenges, clearly. As far as looking at this from the perspective of streaming platforms and broadcasters, it seems like there also are integration challenges. How does this sign language interpretation fit into the infrastructure we already have? Could you walk us through what integrating AI-generated sign language into an existing live or on-demand pipeline actually looks like in terms of ingest, latency, and output formats?
So the core of the system that Signapse built is text input, and then you get a video out of that. That's the core of the idea behind it. So the text that you get could be from captions that you generate anyways, or it could be generated from audio that you can generate as you would any subtitle generated from audio. You generate the subtitles, then you feed these subtitle information into Signapse's system and then they create a video output of that and then you return it.
The difficulties arise on different dimensions. The first one is, if you are in a live setup and you have a live feed that is coming in, you have to feed that into the system here. What if the sign language interpretation takes longer than the live feed? Let's say somebody talks very fast in a live video stream and the sign language interpreter cannot keep up. If you have seen live interpreters in a live setup, like in a book reading or something like that, you might know that people who are talking and seeing that there is a sign language interpreter besides them, will stop and pause until the sign language interpreter has finished translating.
In a live stream that comes from a broadcast, you cannot really do that because the people on stage have no idea that there is a sign language interpretation taking place. So what can you do? You can either take the time it needs and then the sign language interpretation might get longer than the original content, which is in some cases what you want, but in many cases is not what you want. Or you can try to accelerate it and you don't want to have the motions of the sign language interpreter being played at a higher speed.
That's not how a deaf person who needs these sign language translation would consume it. It has to be a certain pace. So there has to be an editorial process in between that takes the spoken word or the text that is in the input and then adopts it to the available time. So there is a constraint that says, "This sentence, this is the information that was spoken in 20 seconds, the sign language must not take longer than these 20 seconds, otherwise the video is delayed." So the biggest challenge is on this linguistic editorial time-constraint dimension.
The other part is practically what G&L does on a daily business: convert all the different formats that we have into one that Signapse understands. So take SRT and RTMP and HLS and DASH and all the different formats, transform that into something that we can then quickly send over to Signapse, get back their video, and do a quality check on the technical terms. Is the resolution right? Are there any artifacts? If not, integrate that into the live feed or into the on-demand translation feed.
Then it mostly boils down in live streaming to the latency that you asked for. And we can bring this down with Signapse currently to something like 12 seconds, or 10 seconds. They're working now to reduce that. And that means that for many use cases in live streaming, we simply delay the video, wait until the sign language is generated, and then sync it back on the video and only then proceed. So there is a small ring buffer that takes the video, waits until the sign language is ready, and only then sends it out. Because for many broadcast-like scenarios, there is no need for ultra low latency. So that's one of the use cases. Wait until the sign language is available, which takes about 10 seconds, and then stream it out.
If, on the other side, we are using it for on-stage live translation, we are also looking into that. So we have a huge display that is as tall as a person. We are showing that display at the IBC. Then you have an audio microphone--for instance, you talk into it, and then you want the most ultra-low latency that generates the code on the other side. Then latency plays a total important role. And so that's where we want to bring this down to six, five, or perhaps four seconds. That would be the goal in the near future.
So I imagine there are also significant localisation challenges with this. And when you move to a new sign language, whether that's one that Signapse has already announced like British or American Sign Language where there are significant differences, or another one like German Sign Language, how much of the underlying pipeline can actually be reused and how much has to be built from scratch for each language?
First of all, the diversity of sign languages is that there are more than 200 worldwide. There is one international one, the International Sign Language, but it's not very popular. It's like Esperanto, which is also an international language that nobody speaks. So there are only very few interpreters who speak the International Sign Language. And the other sign languages are not necessarily along the lines of the spoken languages. So British English and American English are very close to each other. You can easily understand each other. The sign languages are pretty different because historically they have been influenced by different schools of sign languages, mainly dependent on which monk from which school traveled in the 19th or 18th century to which part of the world. So the American Sign Language, to my knowledge--and I'm not an expert on that--is influenced by French Sign Language as is the Irish Sign Language, while the British Sign Language is completely different.
As Germans and Austrians, we can speak to each other easily, but our sign languages are pretty different. So there are mixtures and barriers that you would not necessarily expect in the world. So the process of recording and of generating it, the first is a process that is individual per language because it dives into the specific syntax and grammar that each language has. If you have done one language of the French school, there are some things that you can repeat and adopt, but usually you have to have one linguistic dive-in per language where you completely understand how the grammar is, how it's built up, what the differences are. Then the actual recording which Signapse has come up with and iterated over the three languages that they covered, so that we are now in the third generation of their setup, is pretty easy.
So you have some 4K, 3D cameras from different perspectives. We have a greenscreen behind the sign language interpreter. They sign about 10,000 glosses, which are the basic atoms of the sign language movements and expressions. And then these are recorded, analysed, and normalised, and then checked. So this whole recording process, building up the studio, and traveling to where the respective sign language interpreters are, is pretty repeatable. So the setup is something that Signapse has optimised over the time, that's repeatable. They put it into their models, adopt it, and from there on, you don't have a real difference in interacting with it. So for a customer of our solution with Signapse, the difference between using the American Sign Language, the British, or the German, is just a flag in an API. So that's something that is completely transparent for the end user of the solution. The main difference is at the beginning, diving into the different linguistic aspects of each sign language.
One angle I think would also be interesting to explore here is how these developments are being received in the deaf community. I can imagine there's understandable skepticism within the deaf community about AI avatars, largely as a result of poor attempts in the market in the past. What gave you and the team at G&L confidence that Synapse's models were different, and how did that shape your approach to the project?
We are not sign language speakers, and we don't have deaf people at G&L. But what I know from my own personal experience is, if you are in fields where you really know your way around... For instance, I love Renaissance music. So if you're an expert for Renaissance music, you have a very good antenna if something is wrong. If somebody is talking about this topic and does not know anything about it, you immediately get it. And so I have a huge respect for how the deaf community interacts with any sign language interpreter, be it a human or a machine, because it's their native language. And it was clear to us, after our own failed trials, if we found somebody who would cover this, it would have to be someone who's deeply connected to the deaf community.
And so the nice thing with Synapse is that they check all the boxes. They come from that background. The CEO, Sally Chalk, managed one of the biggest UK sign language interpretation agencies. So she knew what this was about. Then it's academic in its background. It comes from the University of Surrey in the UK, with their image processing capabilities and experience and the linguistic departments behind that. They had a close connection to the deaf community from the beginning. They have deaf people in the company, and two of the corporate stakeholders of Signapse are deaf communities in the UK who are in turn led by deaf people.
Also, their approach when we talked with them about integrating German Sign Language, the cautiousness with which they approached the German market and the momentum that they wanted to build with the community around that, persuaded us that they would be a really good partner who really cares about the community. That said, we are all clear that a human interpreter is better. So it's not a replacement. It's just what we try to achieve is there are not enough human interpreters. We need the human interpreters who are there for the bidirectional communication where they're irreplaceable. But for the rest of the content, sign language generated by AI can be more than good enough to open up more content for the deaf community. That's what we want to achieve. It's not a replacement of human interpreters. And I think it's important to communicate this clearly with the deaf community. And I think Signaps is doing a great job for that.
I understand that G&L and Signapse are going to be sharing a stand at IBC this year, which is starting in just a few days. For anyone who's attending the show, what can they expect to see and experience there? And looking ahead to this autumn and beyond, what's next for the project?
What you will see is, we have a huge, two-metres-high, vertical display where we show the full resolution of the generated video. So we have a demo video where you can see the different languages, and how they are being signed. We have access to all the APIs so that we can do interactive demos on that screen as well so that you can see how quickly it is generated or what options the APIs have that you might want to integrate into your service.
What's next is that Signapse is working hard on getting more languages aboard, so they will definitely expand into different languages. We are very much focusing on perfecting the broadcast integration so that whatever format you have, we find a way in which we easily can interact with Signapse so that the output is perfect, on time, low latency, and stable.
Beyond that, together we are looking into ways in which we can get more emotional variation or perspective variation. So for instance, one request that we often hear is, "Hey, if we have a sports broadcast, it makes a huge difference if the goal is for one team or the other." In some way, there needs to be some emotion included here. So we are looking into how can we capture emotion from the spoken word? There are some ways to do that. How can we transport that? And how can these emotions be expressed in the sign language interpreter that we generate? So those are the things that we look into in the near future. But first of all, we want to launch the German Sign Language. Then we can optimize from there.
Visit G&L and Signapse at IBC 2026
Related Articles
Provenance and authenticity of content have become critical issues as Generative AI-derived content and disinformation flood the digital world. The Coalition from Content Provenance and Authenticity (C2PA) has formed to cryptographically prove content authenticity in today's digital landscape, with an eye to combatting fake content and restoring trust in media. In this interview with Streaming Media's Steve Nathans-Kelly, G&L Systemhaus CEO Alexander Leschinsky discusses the challenges and limitations of C2PA--particularly with live content--and G&L's role in helping broadcasters and media companies implement C2PA standards.
09 Jan 2025
With key streaming services such as Disney+, Amazon, and Netflix trying to drive down production costs across the board, premium content providers have spent considerable time looking at how they can develop or license content which isn't produced in English but can offer global appeal.
26 Aug 2024
Alexander Leschinsky talks data sovereignty, sustainability and the state of the streaming industry.
19 Aug 2022
Companies and Suppliers Mentioned