What is a spoken dialogue system?
Trends and mechanisms of spoken dialogue systems. | SLU
At the Beginning
Hello! My name is Suezugu, and I am writing an article explaining Natural Language Processing (NLP) and Natural Language Generation (NLG)!
In this article, I will explain spoken dialogue systems.
Simply explained, a spoken dialogue system is “A system that understands spoken input and performs appropriate tasks.”
For example, imagine talking to Siri or Alexa.
Alexa, what’s the weather going to be like tomorrow?
Hey Siri, play a storm song.
You can ask about the weather, play a song, or do various other things.
You can ask about the weather, play a song, or do a variety of other things with your voice, such as “What’s the weather tomorrow? and “Hey Siri, play a storm song,” etc. The technology used to process these voice inputs is a spoken dialogue system.
What we will explain in this article
And in recent years, Siri, Alexa, and others have begun to become popular, increasing the demand for spoken dialogue systems. On the other hand, there is not much information on spoken dialogue systems on Japanese websites.
Therefore, this article will provide an easy-to-understand and comprehensive explanation of spoken dialogue systems.
Specifically, we will discuss “Why is it necessary to learn about spoken dialogue systems? and “How does a spoken dialogue system work?”.
Why is it necessary to learn about spoken dialogue systems? The High Potential of Spoken Dialogue Systems
Why is it necessary to learn about spoken dialogue systems, which we will introduce in this article?
The answer to this question can be summed up in one word: Increased demand. In recent years, the use of spoken dialogue systems, and voice assistants such as Alexa and Siri, have become increasingly popular.
Furthermore, since voice input will be used in AR/VR and IoT in the future, the demand is expected to increase even more.
The rapid increase in demand
Let’s take a look at some actual data related to spoken dialogue systems.
According to the latest survey results.
- Nearly one-quarter (24%) of adult Americans own a voice assistant as of spring 2020. This is a 3% increase from 21% in 2019. (National Public Media, 2020)
Voice shopping in particular is on the rise, with sales reaching $1.8 billion in 2018. (Hayllar & Coode, 2018) - In China, more than 1 million orders and payments were placed through Alibaba Genie’s voice shopping feature during Alibaba’s 24h 11.11 shopping festival (Li, 2019, p. 11).
- Thus, we can see that voice assistants are steadily becoming popular in the United States, China, and elsewhere, especially in e-commerce, where they generate significant sales.
It is said that devices using spoken dialogue systems will continue to become more popular, further increasing the demand for this technology.
The Value of Learning Spoken Dialogue Systems Now
As explained so far, the demand for technologies related to spoken dialogue systems is expected to increase.
If you are currently studying data science or NLP and want to specialize in something, spoken dialogue systems are a good option for you. (In fact, more and more people are starting to study data science, making it important to differentiate yourself from others.)
Get Suetsugu’s stories in your inbox
Join Medium for free to get updates from this writer.
So far, we hope you have gotten a sense of the high potential of spoken dialogue systems and the need to learn about them.
In the next chapter, I will explain how spoken dialogue systems work!
How a Spoken Dialogue System Works
As shown in the illustration, a spoken dialogue system mainly consists of the following four steps
- Voice input
Alexa, what is the weather like tomorrow? - Speech recognition (ASR)
Converting speech into text - Language understanding (NLU)
From input sentences, “What does the user want?” from the input sentences - Actions
Display weather, reply, etc.
1 to 4 are collectively called a spoken dialogue system, and 2 and 3 are called SLU (Spoken Language Understanding).
SLU stands for Spoken Language Understanding, which is a technology that “understands input through the human voice” in particular.
This technology is the core of a spoken dialogue system that understands “What does the user want to do? This technology is the core technology for understanding “what the user wants to do” in a spoken dialogue system.
Evolution of Technology in Spoken Dialogue Systems
Up to this point, I have explained how devices using spoken dialogue systems (Alexa, Siri, etc.) are becoming increasingly popular. From here, we will explain the technological development of spoken dialogue systems.
In conclusion, the technology of spoken dialogue systems has developed rapidly in recent years.
The technologies used in spoken dialogue systems, speech recognition (ASR), and language understanding (NLU) have improved considerably over the past few years.
- Improvements in Speech Recognition (ASR) Performance
The error rate for speech recognition has improved to about 1–2%. The human error rate is said to be around 4% and has improved to an accuracy higher than that of humans. (Source) - Improved performance of language understanding (NLU)
There are two measures of NLU performance: “intention comprehension” and “slot filling. Each of these has achieved an accuracy of 98% or better. (Source)
About recent technological innovations
Thus, performance has improved significantly in recent years, and further research is underway to enable spoken dialogue systems to perform more complex tasks.
Currently, most tasks are simple, such as “play music” or “turn off the lights,” but it is likely that more complex commands will become possible in the future.
In recent years, voice assistant-related performance has improved dramatically since around 2018. This is mainly due to BERT, which was developed in 2018. (BERT is an innovative model of NLP)
BERT-based models are producing high accuracy in both speech recognition and language understanding. For speech recognition, w2v-BERT XXL and Stack-Propagation (+BERT) have the best performance as of 2022.
Learn on this site
We will continue to publish explanatory articles on spoken dialogue systems.
Blog URL：https://nlpillustration.tech/?p=2310
We plan to explain everything from theory to technical details in an easy-to-understand manner. Specifically, we plan to explain everything from basic areas such as slot filling and intent detection. We plan to start with the basics, such as slot filling and intent detection.
I will tweet updates as they are posted, so if you are interested in NLP, please follow me on Twitter. (If you have any NLP-related articles you would like to see explained, please send them to me by DM.)
