Creating a SpeechKit Hybrid demo stand

SpeechKit Hybrid enables Yandex SpeechKit speech recognition and synthesis. You can deploy your SpeechKit Hybrid demo stand using Yandex Cloud services through Terraform. This way, you can test recognition and synthesis applications hosted in Docker containers.

Creating a demo stand involves using two machines:

• Local one (in our example, it is Linux-based).
• Virtual one that meets the SpeechKit Hybrid system requirements. This VM runs Docker containers.

Our demo stand uses the Cloud Billing licensing model; this means the info on each request to SpeechKit Hybrid is sent to Yandex Cloud Billing.

To deploy your SpeechKit Hybrid demo stand:

1. Get started with Yandex Cloud.
2. Install additional dependencies.
3. Prepare the SSH keys.
4. Add variables for the Terraform configuration.
5. Create an infrastructure using Terraform.
6. Set up a permanent communication channel with the Yandex Cloud server.
7. Perform load testing for speech recognition and synthesis.

In case of errors, use our debugging guide.

Get started with Yandex CloudGet started with Yandex Cloud

1. Sign up for Yandex Cloud. Signing up is different for individuals and legal entities:

   • How to sign up as an individual
   • How to sign up as a business

2. Go to the management console and log in to Yandex Cloud.

3. Create a directory in the management console. It will contain your resources:

   1. In the management console, select the appropriate cloud in the list on the left.

   2. At the top right, click Create folder.

   3. Enter the folder name. The naming requirements are as follows:

      • The name must be from 3 to 63 characters long.
      • It may contain lowercase Latin letters, numbers, and hyphens.
      • The first character must be a letter and the last character cannot be a hyphen.

   4. (Optional) Enter a description of the folder.

   5. Select Create a default network. This will create a network with subnets in each availability zone. Within this network, a default security group will be created, inside which all network traffic is allowed.

   6. Click Create.

4. Create the sk-hybrid-example service account.

   The service account allows you to flexibly configure access permissions. For more information about the service account, see Service accounts.

5. Assign the following roles to the service account:

   • compute.editor: To create the Yandex Cloud VM.
   • container-registry.images.puller: To work with Docker images in the Yandex Container Registry registry.
   • iam.serviceAccounts.keyAdmin: To create an API key for authorization in Yandex Cloud Billing.

6. Create an API key.

   Save the ID and the secret part of the key. You cannot request them later.

7. Create a registry in Container Registry.

8. Provide the registry ID to the SpeechKit team. The required containers and images will appear in your registry.

Install additional dependenciesInstall additional dependencies

On a local machine:

1. Install the Yandex Cloud command line interface (YC CLI).
2. Authenticate your service account via the YC CLI.
3. Install Terraform.

Prepare a repository with the Terraform configurationPrepare a repository with the Terraform configuration

On a local machine:

1. Clone the repository with the Terraform configuration from which the required infrastructure will be deployed:

   git clone git@github.com:yandex-cloud-examples/yc-speechkit-hybrid-deployment.git
   

2. In the terminal, go to the cloned repository directory.

Prepare the SSH keysPrepare the SSH keys

You will need the SSH keys for authentication when connecting to the Yandex Cloud VM. To prepare them, perform the following steps on the local machine:

1. If you do not have a pair containing a public and private SSH key, create one:

   ssh-keygen -t rsa -f $HOME/.ssh/speechkit_hybrid
   

   After running this command, you will be prompted to enter the password for the private key. If you do not want to provide a password, click Enter.

2. In the directory of the cloned repository, create a symbolic link that points to the public SSH key:

   ln -s ~/.ssh/<key_name>.pub ./keys/ssh-user-id-rsa.pub
   

   This command provides the following:

   • ~/.ssh/<key_name>.pub: File with a public SSH key. If you created the key in the previous step, specify ~/.ssh/speechkit_hybrid.pub.
   • ./keys/ssh-user-id-rsa.pub: Symbolic link. The path is relative to the current repository directory.

Add variables for the Terraform configurationAdd variables for the Terraform configuration

The terraform.tfvars.template file is located in the yc-speechkit-hybrid-deployment repository directory. It is a Terraform template by which environment variables are set. These variables are provided to the YC CLI and Terraform when running commands.

To set variables for the Terraform configuration, perform the following steps on the local machine:

1. Create a copy of the Terraform template in the yc-speechkit-hybrid-deployment repository directory:

   cp ./terraform.tfvars.template ./terraform.tfvars
   

2. Specify the variable values in the terraform.tfvars file:

   • CR_REGISTRY_ID: Container Registry registry ID
   • BILLING_STATIC_API_KEY: Secret part of the API key

3. Optionally, add the NODES_GPU_INTERRUPTIBLE = false variable.

   The Terraform configuration in the repository assumes creating an interruptible VM. You can disable interrupting using the NODES_GPU_INTERRUPTIBLE variable. Its default value is true, and it is specified in the variables.tf file in the yc-speechkit-hybrid-deployment repository.

Create an infrastructure using TerraformCreate an infrastructure using Terraform

The infrastructure required to work with SpeechKit Hybrid is described in the networks.tf and node-deploy.tf files in the yc-speechkit-hybrid-deployment repository. The networks.tf file contains the configuration of the following entities:

• Network
• Subnet
• Internal DNS zone
• Security group

The node-deploy.tf file contains the VM and SpeechKit Hybrid configuration.

Read more about entity configuration on the Terraform website and in the relevant Yandex Cloud service documentation.

To create the infrastructure, perform the following steps on the local machine:

Set up a permanent communication channel with the Yandex Cloud serverSet up a permanent communication channel with the Yandex Cloud server

To work according to the Cloud Billing licensing model, ensure network connectivity between the Yandex Cloud Billing billing.datasphere.yandexcloud.net:443 node and the VM on which the SpeechKit Hybrid service is deployed. To check the node for availability:

1. On the local machine, get the public IP address of the created VM:

   yc compute instance list
   

   The public address will be needed to connect to the VM.

   Result example:

   +-----------+-------------------------------+---------------+---------+-------------+--------------+
   |     ID    |              NAME             |    ZONE ID    | STATUS  | EXTERNAL IP | INTERNAL IP  |
   +-----------+-------------------------------+---------------+---------+-------------+--------------+
   | fhmjvr*** | sk-hybrid-compose-example-*** | ru-central1-a | RUNNING | 158.160.*** | 192.168.***  |
   | ...                                                                                              |
   +-----------+-------------------------------+---------------+---------+-------------+--------------+
   

   The public address is specified in the EXTERNAL IP field.

2. Connect to the VM over SSH:

   ssh <username>@<VM_public_IP_address>
   

   Where <username> is the VM account username.

3. Run this command:

   nc -vz billing.datasphere.yandexcloud.net 443
   

   If the node is available over the network, the command will return the following result:

   Connection to billing.datasphere.yandexcloud.net 443 port [tcp/https] succeeded!
   

Perform load testing for speech recognition and synthesisPerform load testing for speech recognition and synthesis

To check whether the SpeechKit Hybrid test installation is valid and its performance is fine, use Docker containers with the load testing utility for speech recognition and synthesis. These containers are described in the node-deploy.tf file, they were created along with the infrastructure.

To perform load testing:

1. Connect to the VM over SSH.

2. Make sure ports 8080 and 9080 are open to receive client requests:

   telnet <VM_public_address> 8080 && telnet <VM_public_address> 9080
   

3. Run speech recognition:

   docker run --rm --name stt-tools \
      --env ENVOY_HOST=<VM_public_address> \
      --env ENVOY_PORT=8080 \
      --env CONNECTIONS=40 \
      cr.yandex/<registry_ID>/release/tools/stt-tools:0.20
   

   In the command, specify the public IP address of the VM and the ID of the previously created Container Registry registry.

   Where:

   • ENVOY_HOST: IP address of the recognition service.
   • ENVOY_PORT: Port of the recognition service (8080 by default).
   • CONNECTIONS: Number of simultaneously active channels.

4. Run speech synthesis:

   docker run --rm --name tts-tools \
      --network=host \
      --env ENVOY_HOST=<VM_public_address> \
      --env ENVOY_TTS_PORT=9080 \
      --env RPS=20 \
      cr.yandex/<registry_ID>/release/tools/tts-tools:0.20
   

   In the command, specify the public IP address of the VM and the ID of the previously created Container Registry registry.

   Where:

   • ENVOY_HOST: IP address of the speech synthesis service.
   • ENVOY_TTS_PORT: Port of the speech synthesis service (9080 by default).
   • RPS: Number of speech synthesis requests per second.

5. Wait a few minutes while speech recognition and synthesis are performed.

6. Look at the test results in the container logs:

   • docker logs stt-tools: For speech recognition.
   • docker logs tts-tools: For speech synthesis.

   Until the Load finished. Ready to serve requests on 0.0.0.0:17001 line appears in the logs, the speech recognition and synthesis services will not respond to requests. This may take from 2 to 10 minutes.

   Next, the logs will show a message that the Envoy component has started listening to port 8080 for speech recognition and port 9080 for speech synthesis. This means SpeechKit Hybrid is running and ready to serve client requests.

7. Optionally, stop load testing.

   During load testing, the docker run commands will not respond to the Ctrl + C interrupt signals. If you want to stop the containers from running, run the following command:

   • docker stop stt-tools: For speech recognition.
   • docker stop tts-tools: For speech synthesis.