Tilde has worked on spoken language processing since the late 1990s.
Tilde’s researchers has developed a world's first Latvian automatic speech recognition and text-to speech systems. The research result - Latvian automatic speech recognition service -is freely available as a web service. Latvian automatic speech recognition and text-to speech systems are also integrated in Tilde's products.
Driving Force for Development of Speech Technology for Baltic Languages
Tilde continues research on speech recognition by adapting developed technologies for new languages (e.g. Lithuanian) and for specific domains (e.g. medicine, public debates, etc.). The special attention is paid to data sparseness problem that is typical for morphologically rich languages and to novel methods for data acquisition from the web. New application domain - dictation – is also among active research topics. Finally, methods for speech recognition from low quality data are actively researched.
Where it concerns speech synthesis, Tilde’s experience cover different methods, including unit selection, parametric synthesis and recently parametric synthesis with neural networks.