Text to Speech Open Api Infra

This repo helps deploy our TTS services on different infra platforms. We use helm charts to deploy our components.

TTS API Documentation: https://open-speech-ekstep.github.io/tts_model_api/

Deployment Guide: https://open-speech-ekstep.github.io/api_deployment/

Branch	Purpose
azure-deploy	Deployment on Azure AKS
aws-deploy	Deployment on AWS EKS

You can fork the branch closer to your infra and built on top of it. For the details of TTS deployment look at TTS API repo.

infra directories

Directory	Purpose
infra/tts-model-v1	Scripts and helm charts for model pods.
infra/envoy	Scripts and helm charts for envoy to support routing and load distribution across pods.

Below are the details for some files that you need to know to start with this.

app_config.yaml

This is main file to define the cluster of languages you want to have in one single pod. It supports few more parameters as mentioned below.

base_name: tts-model-v1 # base name for the pod.
config:
  - languages:
      - en # languages to be deployed in the pod.
    gpu:
      count: 0
      accelerator: 'nvidia-A100' # (Optional) gpu accelerator parameter.
      CUDA_VISIBLE_DEVICES: "2" # (Optional) index of the GPU exposed.
    replicaCount: 3 # (Optional)
    nodeName: "scn32-100g" # (Optional) node selector name.
  - languages:
      - ne
      - doi
    gpu:
      count: 0
      accelerator: 'nvidia-A100'
      CUDA_VISIBLE_DEVICES: "1"
    nodeName: "scn32-100g"

deploy.py

We have created a python script to dynamically generate all the deployments as the cluster config changes.

Jenkinsfile

Jenkins config file.

Running it on local

python3 deploy.py --namespace <namespace> --api-updated <boolean value> --enable-ingress <enable ingress for envoy admin> --image-name <model image> --image-version <docker image version> --enable-envoy-admin <deploy envoy admin component>

Example:

python3 deploy.py --namespace stage --api-updated true --enable-ingress false --image-name gcr.io/ekstepspeechrecognition/text_to_speech_open_api --image-version 2.1.21 --enable-envoy-admin true

Documentation Reference:

TTS Repo https://github.com/Open-Speech-EkStep/text-to-speech-open-api.

Developer Guide https://open-speech-ekstep.github.io/tts_model_api/