Glossary
Spoken dialogue system
A spoken dialogue system is a technical system that conducts a dialogue with people through spoken language.
On this page
It is the umbrella term from telephony and research; in German-language tenders it appears as “Sprachdialogsystem”. It covers phone menus (IVR), voicebots, voice assistants in cars and AI phone assistants. What they share is the sequence: recognise speech, determine the meaning, choose the next step, output a reply. Where needed, access to data is added, for instance to a customer account or a calendar.
Systems are distinguished by who leads the dialogue. System-led dialogues ask one question after another and expect short answers. Mixed-initiative dialogues allow callers to give several details at once or to change the subject. With large language models, the second form has become the normal case. Tenders and texts from public authorities still mostly use the older term, even when an AI phone assistant is meant. That changes nothing about the requirements: what counts is which dialogues the system is supposed to conduct.
In practice
Anyone planning a spoken dialogue system, or putting one out to tender, does best to describe the dialogues rather than the technology. That means the most frequent requests, the details the system has to collect, the systems it accesses and the cases in which a person takes over. The requirements for recognition, dialogue management and interfaces follow from that description almost by themselves. For acceptance testing, use calls with real wording, including half sentences, questions back and changes of subject.
An example
A city council wants to take pressure off appointment booking at its citizens’ office. The tender document asks for a “Sprachdialogsystem”. What is meant is a system that understands whether someone wants to apply for an identity card or re-register a car, names the documents required and gives out an appointment. A caller says in one sentence: “I need a new ID card, ideally next week in the morning.” A system-led dialogue asks for both details again, one by one; a mixed-initiative dialogue takes them straight from the sentence.
A common misunderstanding
The term is easily equated with one particular type of product, usually the old voice menu. In fact it only states that a system conducts a spoken dialogue, from a simple keypad menu to a freely speaking assistant. Reading the word in a proposal therefore tells you nothing yet about quality. What is revealing is how the system reacts when a caller answers differently from what was planned.
As of: 12 October 2026