Azure AI Speech Voicemail to Text Solution Accelerator

Welcome to the Azure AI Speech Voicemail to Text Solution Accelerator repository! The Azure AI Speech Voicemail to Text Solution Accelerator is a powerful tool that lets you convert voicemails directly to text using speech detection AI!

About this repo

This repository provides a template for setting up the solution accelerator, along with detailed instructions on how to use and customize it to your specific needs.

Overview

This README file documents the process to create a Voicemail to Text Integration Client (V2TIC) project by building a Docker image, deploying it to Minikube, and exposing it as a service.

Documentation

This Readme is designed as a quickstart guide to get you started with the Accelerator quickly. To find more in-depth concepts, visit the index of documents to help you navigate our documentation more easily.

Prerequisites

Option 1: Server-side

• Docker
• Minikube
• kubectl
• Make/Shell/envsubst (available by default on most Ubuntu systems). For Windows, refer to the additional resources section.

Option 2: Client-side

• Python 3.9 (To run sample server listeners and sample clients)
• Pip
  • Open Command Prompt/terminal.
  • Navigate to the location of your azure-ai-speech-voicemail2text folder.
  • Run command pip install -r requirements.txt.

Usage

Clone and checkout the repository.

git clone https://github.com/Azure-Samples/azure-ai-speech-voicemail2text.git

cd azure-ai-speech-voicemail2text

Open the project in Visual Studio Code or any other editor.

Important

In the Makefile in the root folder, you must populate the list of tuple SPEECH_RESOURCES values with your speech key and endpoint URL. The maximum number of tuples in SPEECH_RESOURCES is 2.

Examples:

1. 1 pair of speech key and endpoint. Region failover is not enabled SPEECH_RESOURCES := "[('******************************8b','wss://westus.stt.speech.microsoft.com/speech/universal/v2')]"
2. 2 pairs of speech key and endpoint. Region failover is enabled SPEECH_RESOURCES := "[('******************************8b','wss://westus.stt.speech.microsoft.com/speech/universal/v2'), ('******************************fe','wss://eastus.stt.speech.microsoft.com/speech/universal/v2')]"

Optional: If needed, modify the configmap-file.yaml file to override configuration values under etc/deployments.

Getting Started

• In the scenarios outlined below, Ubuntu 20.04 was used in Linux testing.
• Depending on your installation, the commands below may require sudo.
• Depending on your installation package of python, you may need to input python3.9 instead of python.

Run these commands to build, load and deploy a docker image. All commands must be run in a Git Bash terminal

1. Run cd <path_to_project_parent_folder> where code is checked out.

cd <path_to_project_parent_folder>

2. Run make start_minikube to start Minikube. If it is already running, the status will be displayed. Else, the command will start the cluster. If you need to ensure a fresh minikube container is created, run minikube delete followed by minikube start before this command.

make start_minikube

3. Run make build_image to build the Docker image. Provide a value for <image_name>, else it will default to default_image_datetimestamp. <image_name> must be lowercase.

make build_image image_name=<image-name>

Make a note of the new <image-name>, displayed on screen once the image is created, as a timestamp will always be appended to <image-name>.

4. Run make load_image to load the Docker image into Minikube.

make load_image image_name=<customized-image-name>

5. Run make deploy_pod to deploy and expose the service on Kubernetes. Sample deployment values: https, smtp.

Important

Ensure that the folder name is the same as the deployment name and that the configmap-file and deployment-file yaml files are present in that folder.

make deploy_pod deployment=<name-of-deployment> image_name=<customized-image-name>

6. Run sanity checks on created pod.

Important

You must have executed steps as part of Prerequisites -> Option 2: Client-side. This step must also be performed in a new Git Bash terminal

Make a note of <pod_ip> and <node_port> displayed on screen after execution of make deploy_pod or make quick_start.

a) Start the sample https/smtp listener.

python sample_server_listeners/sample_http_server.py

or

python sample_server_listeners/sample_smtp_server.py

Make a note of <ip_address> on which listener started listening.

b) Add the noted listener <ip_address> as an entry for http or smtp listener server in /etc/hosts file. (for Windows, navigate to C:\Windows\System32\drivers\etc\hosts)

<ip_address> https-listener.v2tic.com smtp-listener.v2tic.com

c) Add the <pod_ip> address as an entry for http or smtp pod server in /etc/hosts file. (for Windows, navigate to C:\Windows\System32\drivers\etc\hosts)

<pod_ip> https-server.v2tic.com smtp-server.v2tic.com

d) Open another terminal and run sample client to send a http or smtp request to the <pod_ip> address and <node_port>

python sample_clients/sample_https_client.py --port <node_port>

or

python sample_clients/sample_smtp_client.py --port <node_port>

e) Check sample client console output for request successfully sent.

f) Check sample listener console output for request successfully transcribed and received.

Optional steps

Run make start_dashboard to open the Kubernetes dashboard.

make start_dashboard

To delete the pod, run make destroy to destroy the deployment, service and configMap for the deployment you select. Sample deployment values: https, smtp.

make destroy_pod deployment=<name-of-deployment>

To clean up artifacts created by make deploy_pod, run the "make clean" command to delete the pod-file.yaml file and configmap-file-instance.yaml file.

make clean deployment=<name-of-deployment>

To enable autoscaling, run make enable_autoscale to enable metrics server and autoscaling on the deployment. Sample deployment values: https, smtp.

make enable_autoscale deployment=<name-of-deployment>

To delete autoscaling, run make destroy_autoscale to destroy autoscaling on the deployment. Sample deployment values: https, smtp.

make destroy_autoscale deployment=<name-of-deployment>

To check and display logs, run make display_pod_logs to display logs from the latest deployed pod. pod_name is displayed on console logs once deployment is successful.

make display_pod_logs pod_name=<name-of-pod>

Run make quick_start to run the end-to-end deployment: building the image from current codebase, starting minikube if it's not already running, loading the image into the minikube cluster and deploying the https or smtp image.

make quick_start deployment=<name-of-deployment>

Additional Resources

• Docker documentation

• Minikube documentation

• kubectl documentation: Commands and tools

• Make: Available on Sourceforge or GNU.org

• WSL

• Cygwin

• Git

Performance Benchmark

Benchmark tests were executed in Azure VM running HTTPS and SMTP servers at a time, using a in-house load test injection tool.

Protocol	VM Size	# Pod	Test Config	LID_MODE	MPM	Metrics	RTF	Converted	Unconverted
HTTPS	D4asV5	1	1 replica , single speech key	AtStartHighAccuracy	240	CPU ~32%, Mem ~19%	0.63	99.99%	0.01%
SMTP (starttls)	D4asV5	1	1 replica , single speech key	AtStartHighAccuracy	240	CPU ~42%, Mem ~15%	0.63	99.99%	0.01%
HTTPS	D4asV5	2	2 replicas , single speech key	AtStartHighAccuracy	360	CPU ~50% , Mem ~19%	0.61 , 0.64	99.96%	0.03%
SMTP (starttls)	D4asV5	2	2 replicas , single speech key	AtStartHighAccuracy	360	CPU ~55% , Mem ~20%	0.42 , 0.42	99.99%	0.01%
HTTPS	D4asV5	2	1 replica , two speech keys	AtStartHighAccuracy	360	CPU ~57% , Mem ~20%	0.65 , 0.65	99.98%	0.01%
SMTP (starttls)	D4asV5	2	1 replica , two speech keys	AtStartHighAccuracy	360	CPU ~61% , Mem ~20%	0.64 , 0.64	99.98%	0.02%

Running V2TIC Regression tests on a local machine

Pre-requisites

• Python 3.9

• Git Client

• Setup ffmpeg (this link will begin the download immediately) on Windows machines.

  [!NOTE] This setup is only needed when you need to run a local server

  1. Download the latest ffmpeg
  2. Unzip the downloaded file to your C: drive
  3. Rename the extracted folder to ffmpeg
  4. Set the environment path variable for ffmpeg by running the following command: setx /m PATH "C:\ffmpeg\bin;%PATH%"
  5. Restart terminal/command prompt
  6. Verify the installation by running: ffmpeg –version

Checkout the branch and install required packages

git clone https://github.com/Azure-Samples/azure-ai-speech-voicemail2text.git
cd azure-ai-speech-voicemail2text
pip install -r requirements.txt

Run test cases

pytest ./tests/api_tests -k regression --capture=tee-sys -s
pytest ./tests/unit_tests

Check HTML Report under reports folder

Note

Currently default speech key and endpoint is set as a F0 pricing tier. In order to change it's value, follow below steps:

1. Navigate to azure-ai-speech-voicemail2text -> etc -> config.properties
2. Modify the value of speech_key and speech_region

Contact

If you have additional questions, please contact the Nuance V2T Migration Team: [email protected], or your Account Representative.

License

This project is licensed under the MIT License.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft’s Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.

Microsoft Open Source Code of Conduct

Please follow the Microsoft Open Source Code of Conduct when using this product.