sabot
AI-driven voice communication with machines
SABOT, the Smart Assistant BOT, enables dialog with all kinds of machines in everyday language. The easiest way to give even complex commands and request specific information. Privacy and security included.
Research project DialoKIM at QUBE
SABOT connects people and machines through AI-supported voice control – experience safe, efficient and flexible machine operation live in real-world applications.
SABOT demo at EuroShop 2023
See SABOT in action in this short demo from EuroShop 2023. SABOT works smoothly even in a noisy environment.
SABOT demo
Let’s take a look at what SABOT can do!
WHY MACHINES THAT TALK
Speech – the most basic human skill
We learn our native language before we even start school and use it throughout our lives to communicate with the world around us and put our thoughts into words – it comes completely naturally.
Machines are becoming more complex; operating them has to be learned and requires attention. Advances in AI now make it possible to control and monitor machines using human language.
Thanks to the AI in SABOT, we no longer have to learn a set of commands or think about how to phrase a question so the machine understands it. It’s enough to say it in clear everyday language. If we aren’t understood with certainty, or if our wording is unclear, contradictory or imprecise, SABOT does what a human listener would do: it asks follow-up questions until a clear, reliable understanding is reached.
Spoken instructions or questions can also be given from a distance, so our eyes and hands can stay on the actual work. This is especially useful when hands are in contact with food, need to stay germ-free when working with patients, or when operating touchscreens with gloves is awkward.
There are several thousand languages in the world, not to mention dialects. To the AI, this makes no difference: any language can be learned.
HOW IT WORKS
Speech
Speaking and listening without a mouth or ears
The acoustic conditions in which dialogs with the machine take place are as varied as the applications themselves. SABOT is a platform that supports various voice inputs and outputs: from a built-in room microphone to noise-canceling headphones, from a lapel microphone to input via phone or smartwatch. The latter is especially practical when communication shouldn’t disturb anyone.
Speech-to-text and text-to-speech conversion and the generation of the artificial voice are handled by online services selected to fit the use case, depending on requirements and cost.
AI that knows its stuff
Machine learning for a lifetime
For SABOT to enter into a dialog, it has to be trained on knowledge about the machine, the process it performs and the conditions it operates in. Above all, safety must always be ensured: misunderstood commands must not cause any damage.
And as with people, knowledge deepens with practice. Machine learning in SABOT is based on this same principle. Every new situation is evaluated and added to the body of experience. This allows series products to keep improving during operation.
SABOT for every machine
SABOT for all kinds of machines
SABOT is a platform suited to retrofitting existing machines. It includes tools for connecting to the machine’s controls and communication interfaces, as well as to IoT platforms, data archives and the IT systems that monitor and control the machines. The same applies to sensors outside the machine itself, so SABOT can also be used for entire plants.
Simulation environments are being developed for SABOT that make it possible to train and test dialogs without the physical machine: this is faster and safer.
Existing machine user interfaces need to be connected to allow smooth, consistent operation: this way, the operator can receive visual feedback or see graphics that are easier to understand than long explanations. The flow of visual and voice operation has to be synchronized. It’s clear: SABOT is not a plug-and-play product, but it makes safe customization easier through a well-understood, efficient approach that SABO Mobile IT, its partners and customers with their own development teams can carry out.
Quick integration
Measurable benefits, fast
The first three questions are: Is voice control possible and practical for a particular machine, and will users accept it? You can try it out with little effort. Prepare to be surprised – now that you have an idea, let’s talk about it!
What’s inside?
The heart of communication with the operator is speech processing. It uses the STT (speech-to-text) component to convert spoken language into text. The text is then processed by a language model, based either on NLP or, more recently, on LLMs, which recognizes the user’s intent.
Based on the context and previous conversations, the system either generates a response or triggers a recognized action. The system also responds to external stimuli, primarily from the machine itself, but data from other sources such as a business intelligence system can also be taken into account.
The operator then receives responses as speech, converted from text using TTS (text-to-speech) components.
MORE INFORMATION
Can SABOT handle our specific vocabulary?
Newer LLMs are trained on large datasets and can recognize most human words, even technical terms. If something is very specific, the models can be fine-tuned with additional terms. Special requirements for SABOT are aligned with the relevant terminology.
Is SABOT secure? Are my private or customer conversations being listened to?
There are several ways to ensure that only the intended speech is processed. SABOT only starts processing speech after recognizing the activation word. Activation words are usually processed locally, and only after positive recognition is the speech processed by STT. Speaker recognition can be adapted to the specific operational requirements, so that usage times match working hours, for example.
How is SABOT integrated into our machine?
SABOT was designed as an open system. It can run as a complete standalone system and, thanks to its multi-service architecture, can be easily integrated into the machine system. A set of SABOT services for managing audio input and output runs on the SABOT hardware. Most SABOT services run on a backend server and perform more complex operations.
What costs are involved with SABOT?
In general, costs depend on the number of operator inputs processed and the number of machines managed in the system. The most cost-intensive services are provided by cloud providers – mainly speech-to-text conversion and the subsequent recognition of meaning. Local options for STT or NLP can also be used, depending on the customer’s requirements.
GET IN TOUCH