A tailored language model for forensic transcriptions

Logo of Finnish institute for health and welfare

Lingsoft’s speech recognition solution accelerates and enhances the work of forensic pathologists, as they no longer need to wait for their dictations to be transcribed. Planner Milla Kemppainen from Finnish Institute for health and welfare explains that the experts in the field can use a custom-ordered, separately tailored language model for forensic transcriptions to shorten the editing process.

Cause-of-death investigations are conducted by seasoned officials in accordance with Finnish law. The Finnish Institute for Health and Welfare (THL) is responsible for conducting forensic autopsies and laboratory tests as part of the cause-of-death investigation process. Thousands of forensic autopsies are carried out in Finland every year, and the process always includes a great deal of documentation in the form of reports and statements.

Before the introduction of speech recognition, the traditional method entailed sending a recording to a forensic pathology assistant for transcription. This meant that the forensic pathologist would have to wait until the assistant had listened to the recording and written it down manually. Today, pathologists can use speech recognition solutions to dictate their reports as they perform their autopsies.

In 2019, THL’s forensic pathology unit purchased Lingsoft’s Speech Service speech recognition solution, which transforms speech into text immediately during the dictation process, allowing the person making the dictation to instantly check their transcription. The solution has been integrated with the forensic pathology unit’s information system and is currently used by every THL unit across Finland. According to Kemppainen, the use of speech recognition has helped markedly decrease the amount of dictations that are sent for manual transcription, allowing THL’s experts to immediately make use of their reports. In addition, many forensic pathologists have praised the solution’s ability to save texts
as phrases that can be used with self-defined speech commands during the dictation process.

In autumn 2021, THL asked Lingsoft to develop a specialised language model and speech recognition solution for its forensic pathology needs. Lingsoft’s previous medical language model had been adequate as-is, but THL’s current specialised language model, which was trained with a state-of-the-art selection of materials, has helped further enhance the accuracy of THL’s automated transcriptions. The tailored model is also able to detect and transcribe the specialised vocabulary and abbreviations used in the field more accurately than ever before, Kemppainen explains.

THL adopted Lingsoft’s language model in February 2022 to mainly positive feedback. Here are some comments provided by medical professionals on the new model:

“It can detect even the trickiest Finnish words, such as sierainaukko (nostril). ”
“So far, the new dictation profile seems to be functioning very well. I haven't been able to stump it yet!”

The forensic pathology unit is an important part of the justice system and represents a vital part of THL’s role as an organisation for public health research and expert services. In addition to its speech recognition solution, Lingsoft has provided THL with a wide range of expert translation and language services.

Read more about our speech recognition solution and the services we provide to health care organisations.