• 2 min read
Yandex devices learn new commands without forgetting old ones
Yandex researchers developed a voice-command update method that adds new phrases without degrading recognition of existing ones.
Image: iXBT
Yandex researchers have presented a method for updating voice-command recognition systems without degrading their ability to understand existing phrases. The work has been accepted for publication at Interspeech 2026, an international conference scheduled to take place in Sydney from September 27 to October 1.
The technology is designed to address catastrophic forgetting: after conventional retraining, voice-controlled devices can become worse at recognizing commands they previously understood. Yandex’s approach adds a separate trainable module dedicated only to new commands, leaving the parameters of the main neural network unchanged. The model grows by less than 10%, while recognition quality for existing phrases is preserved.
Tests on the public Google Speech Commands dataset showed that the method reduced the share of missed new commands to 4.37%, compared with 6.46% for a comparable model. By contrast, conventional full retraining substantially degraded recognition of previously learned commands.
The developers say the approach could be used in smart speakers, household appliances, automotive voice assistants and other voice-controlled devices with limited computing resources.
Recommended reading
Jibo’s successor wants to turn life into AI memories
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via iXBT
