The use of speech technology is becoming increasingly common both in the workplace and in everyday life. At home, voice commands can be used to make phone calls or turn lights on and off, and as more and more household appliances are being connected to the internet, the possibilities for using speech technology are increasing – indeed, smart homes are already being discussed. Speech recognition functions, such as command recognition, also improve accessibility for programs, and even everyday chores with voice-controlled devices.
Speech recognition software recognises speech and transcribes it. The engine has been taught the language and its grammar using large quantities of audio and text. In this process, Finland and other highly inflected languages are considered much more challenging than languages like English with very little word inflection and are thus much easier for speech recognition software to learn. However, the language structure analysis developed by Lingsoft and the speech recognition solution, which is built on it, show that technologies can be developed that fit well with the Finnish language.
The speech recognition software is taught the language using certain types of material. For example, speech recognition tailored for health care is taught using a wide variety of health care-related material so that the machine learns which parts of words typically occur in which context. This kind of specialised speech recognition solution may not work so well in certain other specific fields, but its results are very accurate in the area it has been customised for.
Lingsoft’s speech recognition tools are based on our own language structure analysis, which can be used to teach the software language-specific rules of inflection and word formation. This way, each word does not need to be taught separately; instead, the software learns to recognise different components such as parts of words or word boundaries in hyphenated words and also to identify words in inflected forms.
Speech recognition supporting your work
Speech recognition technology makes it possible, for example, to deliver subtitling on a very tight schedule. Communications often have to be very fast and yet also accessible, and video subtitles or text versions of audio content must be made available without delay. For example, the weather forecast speech recognition model developed for the Swedish national public television broadcaster SVT produced excellent results, with our solution producing more error-free text than that produced by transcriptionists.
There is significant demand for developing speech recognition technologies given how challenging it is to produce real-time subtitling for TV and online broadcasts. Finding solutions for real-time subtitling and other similar challenges represent an important step towards a more accessible world. Lingsoft is tirelessly working to develop solutions, and we are approaching a breakthrough in real-time speech recognition.
Speech recognition provides strong support to human memory, speeds up text production, and frees up working time. For example, typing up the minutes of a meeting is greatly accelerated when speech recognition software creates raw text which then only requires editing. At the same time, this also means that all participants can participate fully in the meeting. Speech recognition is also used in health care for transcription.
Speech recognition facilitates the processing of large audio archives by producing text-based data which is easier to handle and organise. This significantly improves findability. Our findability solutions can also improve data security, as sensitive data, such as personal information, can be located faster and anonymised or deleted in accordance with data protection requirements. At the same time, it is also possible to add metadata to the required data in order to facilitate the workflow.