Promova launches speaking feature based on machine learning technology

Promova, the multitool with a personalized approach to learning languages, launched the “Speaking” feature in the English-to-English course, allowing language learners to practice pronouncing useful phrases and sentences for future conversations. After listening to the example, users can repeat a particular sentence or phrase from the lesson. The neural network will show the accuracy of the pronounced text in percentage ratio. The exercises with such a feature will engage learners to start speaking and overcome the fear of real-life conversations in a new language.  

“We’ve decided to introduce “Speaking” because there were numerous requests in the feedback from our users. They expressed a desire to have some kind of speaking activity to prepare themselves for real-life conversations and dialogues. In addition, we realized that this feature is a great way to memorize new vocabulary and train your brain to respond and react quickly to someone’s speech since it combines different learning styles,” explains Alisa Sihinishyna, Head of Content at Promova. 

The ”Speaking” feature was created using machine learning technology that can recognize different phonemes. To teach the neural network to recognize sounds in various positions, Promova ML engineers train it on vast data of audios such as reading books or dialogues for listening. They use data sets prepared by professional linguists. All audios were recorded by people of various genders, ages, and backgrounds. The only thing that unites them is that they are native English speakers with American and British accents. To voice the sentence which users need to say with natural-sounding human speech, Promova uses text-to-speech software by Amazon and Google.  

“To get the most accurate feedback when practicing “Speaking” on Promova, we have three recommendations for our users. First, stay in a quiet place to decrease the influence of environmental factors. The additional sounds can distort the phonemes, so the neural network will recognize them as incorrect. Second, speak directly to the speaker on your phone or laptop. The closer the user stays, the more accurate feedback they get. Third, do not use earphones or headphones when using the feature. They are not equipped with the best microphones, so the speech recognition can be suppressed,” advises Ivan Seleznov, ML Engineer at Promova. 

Currently, “Speaking” based on machine learning technology is implemented only in English-to-English courses. However, soon users will be able to use it to learn English through all represented languages in the product. The new feature is available for the Promova mobile app for iOS and Android.