All models · Pricing · API docs

whisper-1 API

whisper-large-v3-turbo on our own GPUs · audio → text. WER in our open benchmark (github.com/vakovalskii/stt-ru-benchmark): clean read speech 4.44%, audiobooks 9.89%, spontaneous speech 7.73%

Runs on our own GPUs in Russia.

How to connect: change one line — base_url. Get an sk- key in the dashboard.

curl https://api.neuraldeep.ru/v1/audio/transcriptions \
  -H "Authorization: Bearer $NEURALDEEP_KEY" \
  -F "model=whisper-1" \
  -F "file=@audio.mp3"

https://neuraldeep.ru/en/models/whisper-1