Score the sound the student just missed.
Language schools need learners to speak every day. A teacher cannot sit with every student on every phoneme. The API returns a score per sound, per word and per sentence so the next drill is obvious.
The problem in the language classroom
A letter grade on the whole sentence
Speaking homework comes back pass/fail or a vague accent mark. The learner does not know which sound to fix, so the next attempt is a guess.
Human raters cannot keep daily pace
Two teachers will not mark the same recording the same way. They also cannot listen to every daily drill in a full cohort.
A transcript is not feedback
Speech-to-text writes what was said. It does not tell the learner how close each sound was to the target sentence.
What the API returns
A score on each sound
Phone, word and sentence scores on a 0–100 scale. Colour them in the lesson UI. The retry is the same sentence, not a new one.
Comparable from one attempt to the next
The same prompt, scored the same way, is what makes a daily speaking habit possible. The number has to mean the same thing tomorrow.
The rest of the lesson stays in one product
One key for pronunciation, speech-to-text and speech synthesis. The course does not grow a second audio stack to ship the drill.
Run it: score pronunciation
This is a pronunciation call, not a text label. Upload a short clip of the target sentence and read the scores on this page.
- Hello90
- world82
- overall86
{
"overallScore": 86,
"sentenceScore": 84,
"decodedTranscript": "Hello world",
"words": [
{ "word": "Hello", "score": 90 },
{ "word": "world", "score": 82 }
]
}Sample response from POST /v1/pronunciation/assess/base64. Run the same call below with your audio.
Make your first call now
A scored first call. Word and phoneme scores appear on this page.
Sample audio of the reference sentence is loaded. Run the call, or replace the file.
POST /v1/pronunciation/assess/base64{
"audio": "<base64-wav>",
"text": "Hello world"
}Sample audio is loaded. You can replace it with your own recording of the reference sentence.
Preparing your API key…
Sample result
Hello world
Phoneme scores
All phonemes in the sentence
curl -X POST https://api.brainiall.com/v1/pronunciation/assess/base64 \
-H "Authorization: Bearer $BRAINIALL_API_KEY" \
-H "Content-Type: application/json" \
-d '{"audio":"<base64-wav>","text":"Hello world"}'{
"name": "Brainiall first call",
"baseUrl": "https://api.brainiall.com",
"authorization": "Bearer $BRAINIALL_API_KEY",
"requests": [
{
"id": "pronunciation",
"method": "POST",
"path": "/v1/pronunciation/assess/base64",
"url": "https://api.brainiall.com/v1/pronunciation/assess/base64",
"body": {
"audio": "UklGRmQBAABXQVZFZm10IBAAAAABAAEAQB8AAIA+AAACABAAZGF0YUABAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA==",
"text": "Hello world"
}
},
{
"id": "tts",
"method": "POST",
"path": "/v1/tts/synthesize",
"url": "https://api.brainiall.com/v1/tts/synthesize",
"body": {
"text": "Hello, welcome.",
"voice": "brainiall-aria"
}
},
{
"id": "nlp",
"method": "POST",
"path": "/v1/nlp/toxicity",
"url": "https://api.brainiall.com/v1/nlp/toxicity",
"body": {
"text": "This product is absolutely amazing"
}
}
]
}The same request, ready to paste outside this browser. Replace the placeholder with your key — the key is never shown here.
What changes for the student
They see the sound that missed
A low score on one phoneme is a drill. A red mark on the whole sentence is not.
They repeat the same day
The course is trying to form a speaking habit. The loop is: say it, see the sound, say it again.
The teacher reviews the class, not every file
The sounds the cohort keeps missing surface without re-listening to every recording.
The number, with its source
Phone PCC 0.682 (premium) vs 0.555 between human raters Published on the pronunciation product page, with the test set and the human-agreement baseline.
Score your own audio
Create a key and run the same pronunciation call. The next screen is this job, not a generic label endpoint.
The next screen is the same job: a key, then this call.