About the company
Gazprom Mezhregiongaz Ukhta LLC is a Gazprom company supplying natural gas to consumers in the Komi Republic and Arkhangelsk Oblast in Russia. In 2018, 5.2 billion cubic meters of gas were sold in the company’s area of business with a turnover of 26.2 billion rubles.
The company provides a reliable supply of gas to consumers and helps establish civilized relations in this market, particularly by improving payment structures and discipline. This requires the continuous monitoring of gas consumption patterns, collection of meter readings, and accounting of payments.
Meter readings are collected by phone with strong seasonal peaks. 400,000 calls are processed on average every year. Speech conversion used to take up a lot of time. In 2019, we decided to use automation to streamline the entry of call-in meter readings.
To create a robotic process for collecting gas meter readings, we employed voice recognition and synthesis technologies provided by Yandex SpeechKit. Experts from Gazprom Mezhregiongaz Ukhta implemented the project. The process included two stages:
- Start up and debugging.
- Adding new areas.
It only took a few days to connect to the default Yandex SpeechKit configuration and get used to it. Then it took us a month to write a script and debug the original processing results. After that, we put the workflow into production.
Adding new areas often gives rise to revisions needed to support local specifics. Speech recognition results highly depend on the quality of the connections and uniformity of how data is submitted. In major cities, users speak clearly and distinctly and follow the script. In towns and villages, poor connection quality (noisy phone lines) and speech characteristics (conversations deviating from the script) have to be accounted for.
We started using Yandex SpeechKit in April, and after six months, we have the following percentage of successful recognition:
- In cities: up to 66%.
- In rural areas: 45-50%.
At the moment, 75% of calls are processed, and by the end of the year, we plan to fully robotize the transcription of call recordings. This will fully solve the issue.
What’s key is the time needed to accept meter readings has been cut in half.
The main reasons for migrating to Yandex Speechkit for speech recognition were its ease-of-use and convenience. These let Igor Druzhkov, our lead software engineer, quickly integrate our accounting system with Yandex.Cloud.