BENGALURU: A Bengaluru firm with expertise in conversational artificial intelligence (AI), whose solution delivered last year is also being used by the Indian armed forces on the western and eastern borders, has now developed a new integrated speech recognition-based solution that can help interpret cross-border intercepts along India’s northern and northeastern borders.
The end-to-end voice translation system developed by Gnani.ai uses automatic speech recognition (ASR), machine translation and speech-to-text to convert Mandarin to English and is designed to help armed forces, intelligence agencies and local law enforcement authorities in improving communication systems and giving substantial leeway to the defence forces.
The firm has used more than 8,000 hours of audio data — four major dialects of Mandarin — to train the machine, and the firm’s co-founder and CTO, Ananth Nagaraj, told TOI the tool would transform and improve the communication systems and help strengthen armed forces.
The solution has a wide range of applications that includes cross border intelligence, voice surveillance, monitoring telephone/internet conversations, intercepting radio/satellite communication and to bridge interactions during border meetings and joint exercises.
The firm also said that their tool has features like noise reduction, dialect/accent detection and support for all audio file formats.
“We are in talks with the armed forces and as on date, we are 100% confident of them accepting this given our previous success. You will hear something from them in the next few months. In 2019, we delivered on a significant contract for the armed forces, and our solution is being used on the western and eastern borders. This changed the game for them,” Nagaraj said.
He said that earlier, the armed forces, which had intercepts from various sources, used either manual analysis or keyword-based search technology. Gnani’s solution used AI and provided a complete transcript of millions of records in a matter of hours.
“For instance, someone had to sit and listen to all the records to find keywords like Aloo or Gobi (code for grenade). They would have to listen to all of them and narrow down how many conversations had such keywords, which was time consuming. With our solution giving a complete transcript, they can simply search,” Nagaraj said.
Nagaraj said the firm realised that the armed forces do not have enough expertise in Mandarin compared to the languages used in Pakistan, Bangladesh or Nepal.
“Working with them (armed forces), we understood that in this case, just giving a transcript was not enough as not many understand Mandarin. So we are offering a transcript in various Mandarin dialects, and a complete translation of the same in English,” he said.
On the value of the 2019 contract, Nagaraj said: “I’m not at liberty to disclose that, but I can say that it was significant. We are a team of 50 AI people in Bengaluru and the contract was worth enough to keep us all happy for a few years.”