Assess 13 English and 4 Japanese task types with the real API
Pick a task type and record.
Choose a language and assessment type, check the prompt and record. Results arrive in seconds, down to individual phonemes.
Assessment type
Kana
Selected kana
あ
a
This demo is for evaluating the features. Your recording is sent to DolphinSOE for assessment. Please do not say anything personal or confidential.
17 task types for every use case.
From pronunciation basics to Q&A, retelling and free-talk speaking tests. Every type can be tried in the demo.
Japanese4
- JA / KANAAPI audio limit 210s
Kana
Scores basic, voiced and contracted kana one at a time.
- JA / WORDAPI audio limit 210s
Word
Scores Japanese words by character and phoneme.
- JA / SENTENCEAPI audio limit 400s
Sentence
Scores accuracy, fluency and completeness of Japanese sentences.
- JA / CHAPTERAPI audio limit 400s
Paragraph
Analyses long Japanese passages down to sentences, characters and phonemes.
English13
- EN / PHONEMEAPI audio limit 20s
Phoneme
Scores vowels and consonants phoneme by phoneme.
- EN / WORDAPI audio limit 20s
Word
Scores word accuracy, syllables, phonemes and stress.
Children's model available
- EN / WORDCHECKAPI audio limit 20s
Word misreading
Detects substituted, inserted and missing phonemes in a word.
- EN / SENTENCEAPI audio limit 300s
Sentence
Scores accuracy, fluency and prosody of a whole sentence.
Children's model available
- EN / SENTENCECHECKAPI audio limit 40s
Sentence misreading
Reports extra, wrong and missing words in a sentence.
- EN / CHAPTERAPI audio limit 300s
Paragraph
Analyses long read-aloud passages down to sentences, words and phonemes.
- EN / QAAPI audio limit 60s
Q&A
Listen to a dialogue, then answer the question; content and speech are scored.
- EN / RETELLAPI audio limit 600s
Retell
Scores how well you retell a story you heard, and your speech.
- EN / TOPICAPI audio limit 600s
Picture talk
Scores free speech describing a picture.
- EN / RECITEAPI audio limit 600s
Recitation
Recite a memorised text without looking; scored word by word.
- EN / CHOICEAPI audio limit 60s
Multiple choice
Detects which option you read out.
- EN / ELECTIVEAPI audio limit 60s
Elective reading
Detects which candidate sentence you read and scores it.
- EN / FREEDOMAPI audio limit 600s
Free talk
Scores fluency and accuracy of unscripted speech.
One recording, one full report.
Everything comes back as JSON, ready for your learning app's screens and score history.
- 01
Overall score
overall0–100 rating of overall pronunciation; bands show proficiency.
- 02
Accuracy
accuracyHow close each sound is to the reference; flags misread or unclear sounds.
- 03
Fluency
fluency.overallFew breaks and restarts; how naturally words connect.
- 04
Completeness
integrityHow much of the text was read; reflects missing and extra words.
- 05
Prosody
rhythmEnglish stress, intonation and phrasing, scored per sentence.
- 06
Phonemes & morae
phonemes[]Score per phoneme; Japanese also returns the reading (katakana) and pitch.
- 07
Pauses & speed
fluency.pause · speed · wpmNumber of pauses, speaking speed (slow / normal / fast) and WPM.
- 08
Misreadings & judgement
type · hitAnswer · indexMissing, inserted, substituted and repeated words; choice correctness and which sentence was read.
Token, stream, result: three steps.
- TOKEN
Get a token on your server
Sign with your AppID and Secret to issue a token. The Secret never reaches the browser.
- STREAM
Send audio over WebSocket
Stream 16 kHz, 16-bit PCM while recording. The task type and reference text go in the start message.
- RESULT
Receive the assessment as JSON
As soon as recording stops, the overall score, metrics, phonemes and misreadings come back.
Bring this assessment to your service.
Complete the form to request a free API key, which will be sent by email.
Get a free API key
Complete the form to request a free API key, which will be sent by email.
Open the request form in a new window