Ekinox Multilingual Pronunciation Assessment SaaS API

The Ekinox Multilingual Pronunciation Assessment Engine is a full-stack solution for autonomous language learning. Beyond a standard ML model, it combines audio-to-phoneme conversion with a business logic layer that handles pronunciation scoring, fault classification, utterance acceptance, and JSON diagnostic composition. Delivered as a managed SaaS API on AWS Marketplace, it requires no infrastructure to deploy or manage - subscribe and begin sending requests immediately.

See it in action: Try the interactive demo at pronunciation.ekinox.io  to experience phoneme-level scoring and instant feedback firsthand.

• Full-Stack Assessment Engine: Incorporates a dedicated business logic layer for pronunciation scoring, fault classification, and utterance acceptance - all accessible through a single SaaS API endpoint with no deployment or configuration required.
• Phoneme-Level Feedback: Delivers granular scoring at the phoneme level, enabling learners to identify and correct specific pronunciation faults with precision.
• Real-Time Bidirectional Streaming: Each word is confirmed live as the learner reads it.
• Comprehensive JSON Diagnostics: Each assessment returns structured results including overall scores, accepted utterances, and categorized fault details - giving your application the data it needs to drive learner improvement.
• Usage-Based Metering: Costs are metered on actual audio seconds processed, so you pay in proportion to real usage, with no capacity to provision.

• French
• Arabic
• More languages planned

Once subscribed through AWS Marketplace, you send audio input along with the target utterance to the SaaS endpoint and receive a detailed JSON response. The ML model converts audio to phoneme sequences. The business logic layer then scores accuracy, classifies faults, determines utterance acceptance, and composes the full diagnostic output - all in real time, with no infrastructure for you to manage. Live sessions stream audio over WebSocket and confirm each word as it is read.

Audio you send the API is used only to answer your request. It is held only while the request or live session runs, then deleted: it never reaches a database or a bucket, and is never used to train models. Designed for children's recordings, under 13 included: you obtain any consent the law requires. The service and the data it holds (subscription records, usage figures, API key hashes, security logs) are in AWS eu-west-1 (Ireland). The only data sent outside it is your license identifier and the seconds of audio used each hour, to AWS Marketplace metering in the United States, so that AWS can bill you.

Full details: https://pronunciation.ekinox.io/data-handling  Recordings of children: https://pronunciation.ekinox.io/data-handling#children  Security and vulnerability reports: https://pronunciation.ekinox.io/security 

• EdTech Platforms: Add automated pronunciation practice to language-learning applications. Learners read or repeat a phrase and get immediate feedback on which words and sounds to work on, with scores they can follow from one session to the next. Suited to schools and children's apps.

• Corporate Training: Support language development across your organization. Employees practise at their own pace and see the specific sounds they mispronounce, so trainers can spend their time where it helps most.

• Call Centers: Help agents work on their pronunciation as part of coaching. Phoneme-level diagnostics give coaches concrete, comparable material to discuss with each agent.

Note on EU AI Act: Each result compares the speech with the expected text: whether the utterance is accepted, a score in percent, and the faults found, down to each word and phoneme. Supported languages are French and Arabic. Scores are feedback to help people improve. They are not designed or validated for grading, admission to education, hiring, performance evaluation or any other decision about a person. Whether a deployment is high-risk under Annex III of the EU AI Act (Regulation (EU) 2024/1689) depends on how it is used. If yours is, conformity assessment, human oversight and informing end users are your responsibility.

This listing uses usage-based metering on actual audio seconds processed, with no capacity to provision: the service adds instances as load grows. Live sessions run up to a set number at once per subscription; write to us if you need more. A free trial is available so you can evaluate the engine before committing.