EnglishCentral IntelliSpeech℠ Technology
EnglishCentral’s “special sauce” has always been our internally developed IntelliSpeech℠ assessment technology. Teachers appreciate how IntelliSpeech℠ motivates students to speak outside of the classroom. Students appreciate the instant feedback and the game dynamic of trying to improve their speaking scores.
EnglishCentral’s IntelliSpeech℠ is trained on hundreds of millions of lines of speech from students from over 100 countries, using the latest Machine Learning techniques. It assesses learners’ speaking ability across the following dimensions:
Pronunciation measures the acoustical qualities of students’ speech over 64,000 possible triphones (combinations of phonemes), stress and prosodics.
Fluency is still based on duration and pause rate of the speech.
Completion is still based on whether the user speaks all words, or drops words.
Error Types
As the learner speaks, Intellispeech℠ provides feedback according to the following types of errors:
Line Score
Learners get a line score between 0 and 100 points for each line spoken. Learners lose points for each error made. The amount of points learners lose per error is weighted based on the number of words in the line. For shorter lines, learners lose more points per error than for longer lines.
Only the final version of each line spoken counts towards the users final score on the Video Grade. So, if a line is repeated several times, only the last version counts towards the Video Grade.
Video Grades
The video grade is the cumulative measure (i.e. the weighted average) of the line scores received by the learner from speaking lines in the video.
Intellispeech℠ computes a percentile relative to other learners to produce a video grade. For instance, if we determine that the speech for the video was 75% better than other learners, the learner would get a “B+” as a video grade.
Percentiles are mapped to grades as follows:
Pronunciation Center
The Pronunciation Center (“Pron Center”) on EnglishCental is where students’ feedback and progress on their pronunciation is tracked and where they can find exercises to help remediate their pronunciation challenges.
At the heart of the Pron Center are the 4 steps of our Pronunciation Learning Cycle:
Pronunciation Courses
IntelliSpeech℠ has analyzed our users English pronunciation over hundreds of millions of recorded utterances and identified the most common problematic sounds of speakers for each native language region. Based on this analysis and speech data, we have designed pronunciation courses — Top 10 Challenges — where learners can focus on the most challenging Sounds for speakers from their native language, using authentic videos. We currently have customized pronunciation courses for: Japanese, Korean, Spanish, Portuguese, Turkish, French, Vietnamese, and Arabic.
Our Speech Database
For the last 10 years, our learners have provided us a trove of data (over 600,000,000 speech utterances from over 100 countries) on how they learn English.
Our reference models are also training on large amounts of data collected from native speakers speaking the authentic speech from our videos (as opposed to many other corpora which may contain artificial “read speech”).
We use this data to create a “Machine Learning Loop” that combines this large data set with feedback from a team of over 600 trained professional English teachers who have analyzed student speaking ability in over 500,000 1-on-1 live sessions. The result is the “machine” (our online self-study platform) learns from the feedback teachers provide, and at the same time, teachers learn from the feedback the machine provides on learners’ strengths and weaknesses.