Transcribe voicemails to text

In this example project, we will convert voicemails to text using the sipgate.io node library and αcephei Vosk.

What is sipgate.io?

sipgate.io is a collection of APIs, which enables sipgate's customers to build flexible integrations matching their individual needs. Among other things, it provides interfaces for sending and receiving text messages or faxes, monitoring the call history, as well as initiating and manipulating calls. In this tutorial, we will use sipgate.io's history API to read incoming voicemails from your sipgate account.

In this example

The script in this repository pulls the latest voicemail from your sipgate account history and converts it to text using the Vosk speech-to-text model.

Prerequisite: You need npm and Node.js installed on your machine. You also need ffmpeg installed from your package manager.

Getting started

To launch this example, navigate to a directory where you want the example service to be stored. In a terminal, you can clone this repository from GitHub and install the required dependencies using npm install.

git clone https://github.com/sipgate-io/io-labs-speech-to-text.git
cd io-labs-speech-to-text
npm install

To get access to the transcription model, visit the αcephei Vosk website and download a model for your desired language (e.g. vosk-model-small-de-0.15). You can choose between different model sizes depending on your desired accuracy (large models are more accurate than smaller ones).

Extract the files in the .zip and move them to the model folder in the git repository. Your directory structure should look something like this:

├── model
│   ├── am
│   │   └── final.mdl
│   ├── conf
│   │   ├── mfcc.conf
│   │   └── model.conf
│   ├── COPYING
│   ├── graph
│   │   ├── disambig_tid.int
│   │   ├── Gr.fst
│   │   ├── HCLr.fst
│   │   └── phones
│   │       └── word_boundary.int
│   ├── ivector
│   │   ├── final.dubm
│   │   ├── final.ie
│   │   ├── final.mat
│   │   ├── global_cmvn.stats
│   │   ├── online_cmvn.conf
│   │   └── splice.conf
│   └── README
├── package.json
├── package-lock.json
├── README.md
└── src
    ├── main.js
    └── speechToTextEngine.js

7 directories, 20 files

Execution

Make sure to set the credentials of your sipgate account (token and token ID with scope: history, see Personal Access Token documentation on sipgate.io) either in a .env file or by providing them as temporary environment variables at program execution:

SIPGATE_TOKEN=<token> \
SIPGATE_TOKEN_ID=<tokenId> \
npm start

You should now be able to convert your voicemails to text.