LanguaMetrics, Inc. is a world leader in the AI powered analysis of both Children’s and Adult’s speech.
Join us on our voice-powered journey from phonics to fluency to understandability! We have helped millions of learners—both young and old, from Beijing to Bloomington—improve their English reading, their speaking, and help qualify their English in the workforce.
Follow our decade-long timeline of producing cutting-edge speech powered products for major companies and institutions around the world.
What is the LanguaMetrics difference?
Bespoke Artificial Intelligence
We customize our AI models to meet the needs of each application, in each market. This takes time and effort. We do not license other organizations’ technology; we roll our own. Our data scientists are constantly adapting our base models into product-specific models for a wide variety of markets.
Human Rating
The purpose of AI models is to transform human intelligence into machine intelligence. To do this with precision it is necessary to have a machine learning operations workflow, MLOPS. For a decade, LanguaMetrics has maintained a constant pipeline of classified child and adult audio recordings and hand-scored by our trained raters, as well as our partners' raters.
Machine Learning
Our data science team takes in the audio files that have been rated by our human raters to begin the model building and validation process. Our data scientists strive to constantly improve the overall accuracy of our models and quantify the results in industry standard terminology such as Precision, Recall, F1, false positives and false negatives. In AI, accuracy must be quantified across several measures.
Digital Signal Processing
Our audio and electronic engineers are pioneers in deciphering the speech signals. We build proprietary AI models to measure signal features related to background noise so that we can identify recordings that should not be scored because the results could be unfair. We use our signal expertise to expose what are known as “features” so that our data scientists are able to detect patterns that otherwise would not be visible in the data.
Phoneme Level Scoring
All results at the application level depend on our phoneme level scoring capabilities. This can cause information overload if not properly managed. In all our products across the globe, we do analysis at the phoneme level, assigning a numeric score to the utterance's pronunciation accuracy. With these results we have developed an extensive inventory of proprietary algorithms to make scoring and feedback decisions at the word and sentence level based on the phoneme level data
Accents, Equity, and Diversity
Off-the-shelf AI models and speech recognition libraries may be biased because they are based on the most readily available speech data from the internet, which is not sufficiently ethnically nor geographically diverse. Our speech data is collected from the groups of users who use our products, and we carefully monitor its diversity of accent, ethnicity, and gender. Our AI models are held to the highest standards in education and the workforce to ensure scoring results are unbiased and fair to an increasingly diverse global community.
Subject Matter Expertise
Our subject matter experts ensure that the output from our artificial intelligence and speech scoring algorithms is fit for purpose. This ensures our products are not solutions in search of problems to solve. We start with the specific challenge and then custom craft each solution. Our solutions range from the Education space to the Call Center industry and beyond.
