Speech-Enabled Commerce

The recent study "Germany, your speech recognition assistant"* comes to what may be a surprising conclusion for many: speech recognition is in use for more than just playing music, listening to news and retrieving the weather forecast. An increasing number of people are using speech recognition when shopping.

It’s time to integrate speech recognition into online stores. Consumers love simplicity and convenience – not least because they can easily shop from the sofa using a smartphone or tablet while keeping up with their banking. Speech recognition assistants such as Alexa, Siri, Cortana or Bixby - to name the most popular - take digital convenience to a new level. Tedious typing of a search query is yesterday, speaking a search query is today. The main reason for this is ease of use: people don’t want to have to type, switching instead to the comfort of hands-free communications. With voice control users can do several things at the same time.

According to estimates, around 100 million networked speakers were in use worldwide by the end of 2018. By 2022, the number of smart speakers in use is expected to rise to 300 million. In the light of these impressive figures, it would be remiss of online shop operators not to jump on the bandwagon. After all, speech recognition assistants have many advantages in e-commerce:

• Users can make complex enquiries, using voice command to simply place the desired product in the shopping basket and easily query the order status.
• Speech recognition assistants can specifically recommend products that other users have also purchased in combination with a certain product (exploiting upselling and cross-selling potential).

In principle, the possibilities of voice commerce are almost unlimited. Theoretically, exactly the same set of online shop functions can be mapped using a voice-guided assistant,  as can be navigated via the keyboard. The added value or performance of the assistant depends on the algorithm behind the application. In practice, the process works in several stages. Firstly, the received audio signal is converted into phonemes, i.e. into the smallest linguistic units that discriminate between meanings, from which words and sentences are then created. In contrast to classical methods for automated speech recognition (ASR), artificial intelligence (AI) methods are generally used today to improve the recognition rate.

The business logic then determines the user's wish and converts the spoken request into traditional text-based commands to the backend. In a service-oriented architecture, this is usually done by calling web services. The result of the requests is then converted back into speech using TTS algorithms (Text-to-Speech).

Speech-enabled commerce stands and falls with product features. As well as an up-to-date and consistent database, shop operators need to pay the greatest attention to the product descriptions. The language assistant can only find the product it is looking for if it can access an appropriate database. For this purpose, it must be connected to relevant third-party systems: Product Information Management Systems (PIM) or Order Management Systems (OMS).

In order to be able to process spoken search queries in a targeted manner, the right level of semantic understanding is required on the technological side. In this area, however, development is still in its infancy – this is why the results of a complex language-based search query still often fail to meet the user expectations. In contrast to the keywords input for a text-based search, users speak in complete sentences for  language-based search. In order for a voice-recognition assistant to understand "untrained" sentences and deliver an appropriate result, a complex interplay of different algorithms, dictionaries and artificial intelligence is required. In this area, the future will show what is possible and what is practical.