LingoLoop Allosaurus service
This companion service performs experimental acoustic pronunciation assessment.
It converts uploaded microphone audio to mono WAV, recognizes phones with
Allosaurus, generates reference phones with eSpeak NG, aligns the two phone
sequences, and returns accuracy, completeness, fluency, and phone-level hints.
Run with Docker
1docker build -t lingoloop-allosaurus .
2docker run --rm -p 7860:7860 lingoloop-allosaurus
Set the website runtime secret ALLOSAURUS_SERVICE_URL to the public HTTPS URL
of this service.
Run on Render Free
Create a Web Service from this directory's public Git repository and apply the
included render.yaml Blueprint. Keep the payment method empty so exceeding a
free usage allowance suspends the service instead of creating a charge. Free
services sleep after 15 minutes without traffic, so the first pronunciation
request can take longer.
Set the same random value as ALLOSAURUS_SHARED_SECRET on the service and
ALLOSAURUS_SERVICE_KEY on the website so the assessment endpoint cannot be
used anonymously outside LingoLoop.
The universal model is experimental for all languages other than its dedicated
English model. Scores must not be presented as clinical or certification-grade
assessment.
