Recommendations for pronunciation activities

It is important to understand how artificial intelligence (AI) tools work before creating exercises of this type. Otherwise, the activity created may generate frustration for the teacher and students.

By Yanick Demers

Date of Creation: 2020-10-21

Last Update: 2025-08-29

How does speech recognition work?

Speech recognition is a computer technique that analyzes the human voice captured by a microphone and transcribes it into machine-readable text. In fact, the machine will try to guess correctly what the person is trying to say, by analyzing it and comparing it with a large number of recordings.

The score (%) given to the learner/student is not an assessment of the quality of the pronunciation. The AI gives feedback on the interpretation of the recording, in other words, I understood that the pronounced words were close enough to the words, to the written sentence. The AI works like a human being, if you say "Hello" with a bad accent, I can understand that you said it. And for someone learning a language, this is the first step to fluency.

Here are some tips to help you achieve your learning goals with the SmartClass Hub pronunciation activity.

Pronunciation activity

Provide a context to improve the chances of success

It is easier to be understood if you provide context. Let's say I only say the word "four" in English. It is difficult for a machine to tell the difference between "four" and "for". We suggest that you use phrases, rather than single words.

The AI is quite well developed, but it is not a human brain ;-)

Avoid overly specialized vocabulary

The use of technical terms can be difficult to interpret by a machine. Imagine speaking to someone who is new to a cutting-edge field - they may have trouble understanding you!

Be careful with special characters and punctuation

We use Microsoft for our AI capabilities. However, whatever they decided as the rules for training the AI and whatever input they gave to the knowledge base for it, is what our system will mimic. So these aspects are difficult to understand perfectly as we didn't create "the rules" behind the AI.

In our own content activities, we have these instructions for the students:

(Note: For this activity, artificial intelligence is used to score your pronunciation. Sometimes compound or hyphenated words had to be split up or special characters taken out due to the AI not being able to recognize context or language perfectly.)

Example sentence 1: The check-in line was very long. But the AI writes it out after you try to say it as "The check in line was very long." Then simply take out the - because sometimes it will count it as incorrect then.

Take notice of times and numbers

Follow the rule of spelling out the number 0-9, but then type the numeral after 10. However, there does not seem to be a perfect pattern. We see this in one of our activities using ordinal numbers. We have to use 1st, but then spell out second and third. Then "4th" is good, but "fifth" needs to be spelled out.

Example sentence: It is 3:00. But the AI writes it out as "It is 3 o'clock." Change the sentence to this then.

Practice vs. test

Robotel does not recommend using speech recognition to formally assess students. There are too many parameters that can affect the outcome, including the quality of the recording. For assessment, it is best to use an activity model such as an audio or video recording.

The main objective of the pronunciation exercise is to make students practice their pronunciation in a fun way. The note is a tool to motivate them to try as many times as possible.

Check the feasibility of the activity yourself

Finally, it is important to test the activity before sending it to students. To do this, simply preview and run the activity through before sharing to see:

1: that the AI is picking up the sentence correctly. If not, then tweak the sentence. (I always say, if you, as a teacher, cannot get 100%, then you should change the sentence)

2: what the AI is writing out and changing it to match.

This can yield some fun results, so bear with a developing technology.