From voice recorder to article in just two clicks!
Spoken-language recordings, such as from in-depth interviews or focus group discussions, are frequently used in research, particularly within the social sciences and humanities. A tool capable of easily converting spoken language into text directly within a web browser, without requiring the installation of new software, can be very useful for further processing these data.
UWebASR was developed specifically for research and non-commercial purposes, handles recordings of varying quality and length, and is also capable of recognizing historical speech. These aspects make it a tool applicable to a wide range of research fields in the social sciences and humanities.
The development was carried out by the Department of Cybernetics and the NTIS Research Centre at the Faculty of Applied Sciences, University of West Bohemia in Plzeň, as part of the LINDAT/CLARIAH-CZ research infrastructure.
Using the tool
Using the tool is very intuitive; first, it is necessary to select the language and the recognition model. The tool defaults to Czech, with other languages also available: English, Slovak, German, and others. The tool is equipped with instructions regarding the selection of a suitable model for language recognition.
Next, you need to upload the recording to be converted. The tool supports commonly used formats: MP3, MP4, and others. The tool then automatically begins processing the uploaded recording, and within a few minutes, the transcribed text appears in the browser with color-coded confidence scores on a scale from 0 to 1. This makes it possible to immediately check parts that were harder for the tool to recognize. The resulting text can be downloaded in an open .txt format and further processed in any other tool.
It is also possible to use the UWebASR Skill for agentic AI.
Data Protection
Strict privacy rules apply to all listed recognition models. The models retain only anonymous session statistics. Audio recordings and generated transcripts are processed exclusively in-memory during runtime and are immediately deleted upon session termination, without being archived or used for any other purpose.
The underlying platform runs exclusively on physical servers located in the data center of the Faculty of Applied Sciences (Technická 8, Plzeň, Czech Republic). Audio, transcripts, and other recognition data are not shared outside this infrastructure.
For further information or inquiries, a contact person is listed on the platform's website: Jan Švec (honzas@fav.zcu.cz).
The tool and further information about it are available at the link: UWebASR