This section is meant to get you started using Juzu Server via a Docker image.
The SDK communicates with a Juzu Server instance using gRPC. Cobalt distributes a docker image that contains the juzusvr binary and model files. The image can also include Cubic Server as well to get diarization and transcription results together.
Contact Cobalt to get a link to the image file in AWS S3. This link will expire in two weeks, so be sure to download the file to your own server.
Download with the AWS CLI if you have it, or with curl:
URL="the url sent by Cobalt"
IMAGE_NAME="name you want to give the file (should end with the same extension as the url, usually bz2)"
curl $URL -L -o $IMAGE_NAME
Load the docker image
docker load < $IMAGE_NAME
This will output the name of the image (e.g. juzusvr-demo-en_us-16).
Start the juzu service
docker run -p 2727:2727 -p 8080:8080 --name cobalt-juzu juzusvr-demo-en_us-16
That will start a docker container with juzusvr listening for gRPC commands on port 2727 and http requests on 8080, and will stream the log to stdout. (You can replace --name cobalt-juzu with whatever name you want. That just provides a way to refer back to the currently running container.)
If the image provided also includes cubicsvr, then adding the arguments -p CUBICSVR_GRPC_PORT:2728 -p CUBICSVR_HTTP_PORT:8081 to the docker run command above will enable you to access cubicsvr directly via CUBICSVR_GRPC_PORT and CUBICSVR_HTTP_PORT for gRPC and HTTP requests respectively. This is useful when you want to only obtain transcription in “true” streaming mode and diarization is not required.
Verify the service is running by calling
curl http://localhost:8080/api/version
juzusvr - binary for performing Speaker Diarization
diarization models - models for extracting and scoring speaker embedding vectors
If configured to include Cubic: