We used Microsoft Azure as the primary cloud platform to host and scale the web portal. It provided scalable infrastructure, low-latency processing, and easy integration with speech and AI services that we needed for real-time pronunciation analysis.
Discover what’s next for AI in healthcare in 2026 - Get Access to the Full Report
Client Background
Our EdTech client specializes in English language coaching for non-native speakers through guided exercises and technology-enabled instruction. They target students who need accent and fluency training in English language across academic and professional environments.
Country
Japan
Industry
EdTech
Time & Resource Invested
1000+ Hours, 5+ Experts
Project Duration
1 Year 11 Months
Problem Statement
The client needed to build a web portal that helps students improve their English pronunciation and accent through real-time, personalized feedback. Existing learning methods could demonstrate correct speech but lacked precise, real-time evaluation. The requirement was for an AI English learning solution that could accurately assess spoken inputs and offer clear demonstrations of correct speech alongside accurate evaluation of their attempts.
Project Overview
After assessing how non-native speakers practice pronunciation and how automated feedback mechanisms would help learners self-correct, our AI development team designed and built a secure AI eLearning software that visually shows correct pronunciation, listens to student speech in real time, and responds with clear, immediate feedback.
The web portal was built using Azure Avatar for lifelike visual pronunciation models. We paired it with Google Speech-to-Text for accurate transcription and Azure Speech Pronunciation Assessment for feedback on accuracy, fluency, and grammar. It’s all in all an interactive eLearning platform where students see proper lip movements, hear examples, and get instant scores on their learning efforts.
Client Feedback
Our learners come from language backgrounds very different from English. Considering that, the platform had to be nuanced to those criteria, and the solution addressed learning challenges with the level of detail we needed.
Naoki Hashimoto
Director
“It was a demanding project because there was no one-size-fits-all model that would work here. Regional accents vary widely, and English phonetics often clash with native speech patterns. We had to test a lot and retune constantly to make sure that the whole learning process stayed accurate and genuinely helpful for students.”
Ajay Ojha
Chief Architect at Radixweb
Key Challenges
- Real-Time Assessment: Student speech inputs varied widely in accents, speeds, and background noise, which made 100% accurate transcription and pronunciation scoring difficult without false positives.
- Visual Demo Synchronization: Creating lifelike avatar lip-sync and gestures that matched diverse English phonemes was technically demanding in multiple dialects.
- Feedback Clarity: Delivering actionable, non-technical feedback on phoneme-level fluency for thousands of simultaneous users could strain processing limits.
- Cross-Device Performance: Low-latency speech processing and avatar rendering needed to work smoothly on budget smartphones used by students in emerging markets.
- Inconsistent Audio Quality: Learners used different devices and environments, so the English pronunciation training software had to perform despite background noise or low-quality microphones.
Solutions We Delivered
Avatar-Based Pronunciation Modeling
The AI English learning platform uses 3D visual avatars to show phoneme-level pronunciation, with exact lip movements and facial expressions for correct English sounds, so that students can mimic naturally.
Speech Capture and Processing Pipeline
Spoken input is captured and processed in near real time. There’s minimal delay between user speech and system response, which has proved to be critical for effective pronunciation practice.
Pronunciation Scoring
The eLearning portal grades students’ pronunciation accuracy at phoneme, word, and sentence levels, plus fluency and prosody metrics. Students get proper improvement targets.
Personalized Feedback Loops
Instant reports are generated to highlight specific errors, such as vowel distortion or intonation, with simple suggestions and progress tracking over multiple sessions.
Multi-Modal Learning Interface
Avatar demos, audio playback, and scored student recordings are all visible on one screen. The AI language learning platform caters to visual, auditory, and kinesthetic learners.
Scalable Cloud-Based Architecture
The portal is built on Azure cloud architecture. It can handle concurrent users, varied devices, and growing usage without degradation in speech analysis performance.
Validate Your AI Use Case with Us
Run a scoped project with an experienced team of engineers. Get a working prototype, architecture notes, and cost-to-scale estimates within your desirable timeline.
Technology Stack
Microsoft Azure
Google Cloud
Google Cloud was chosen to support high-accuracy speech recognition. Using its APIs, we enabled consistent transcription quality for diverse accents and speaking styles.
Azure Avatar
Azure Avatar was used to visually demonstrate correct pronunciation and articulation. It helped learners understand mouth movements, stress, and sound formation, making pronunciation guidance more intuitive and effective than audio-only instruction.
Google Speech-to-Text
Google Speech-to-Text handled real-time speech transcription with 99%+ accuracy on the AI-driven learning management platform. Our EdTech development team used it to transcribe student speech despite heavy accents or noise. It worked as the foundation for pronunciation verification and analysis.
Azure Speech Pronunciation Assessment
Azure Speech Pronunciation Assessment was used to evaluate pronunciation, fluency, and grammar. The AI eLearning software delivered real-time scoring for learners to identify specific issues and improve through immediate corrective feedback.
Client Benefits
There were measurable benefits following the implementation of ML modules in the eLearning platform. Student engagement increased, so did accessibility scope, user retention rate, and automation capabilities.
Enhanced Learner Engagement
As we implemented real-time feedback and visual pronunciation guidance in the eLearning platform, it noticeably increased active participation during practice sessions. Average session durations grew from roughly 30 minutes to over 50 minutes per learner. Repeat practice attempts per lesson also increased, which indicated higher engagement rather than passive consuming of content.
Better Accessibility for Diverse Learning Styles
With visual demonstrations, audio playback, and textual feedback, the AI-enabled eLearning platform supported 15+ dialects without additional configuration. Learners who struggled with audio-only lessons showed faster improvement. The intelligent language learning solution also reduced dependency on instructor-led sessions. Learning became accessible in different time zones and schedules.
Increased User Retention and Satisfaction
Satisfaction scores reached 4.6 out of 5 following 10-week pilots with 500+ learners. Students liked the instant feedback and visual demos for making practice sessions feel rewarding rather than repetitive. This high rating came from diverse non-native speakers, which actually confirmed the portal's practical value in real classrooms.
Reduced Human Intervention
Automated pronunciation and grammar assessment reduced the need for manual evaluation by instructors. Instructors saved an estimated 22+ hours per week that were previously spent reviewing recordings and providing basic feedback. The AI-powered eLearning portal boosted overall instructional efficiency without increasing operational headcounts.
Work with a Dedicated Engineering Team
Engage a cross-functional team of product engineers, designers, and AI specialists with an average 8+ years of domain experience. Onboarding within 7 to 10 working days.