Text-to-speech (TTS) technology, long established for languages like English and German, is now available for Luxembourgish, marking a significant milestone for the language.
In recent years, Luxembourgish has made notable strides in this field, and the Centre for the Luxembourgish Language (ZLS) has taken a major step forward with a new programme that converts written text into spoken language.
While TTS has been widely used for other languages, its implementation for Luxembourgish posed unique challenges. Following the launch of “Schreifmaschinn” (“typewriter”, lit. “writing machine”) in 2022, the ZLS has now integrated the programme with “Liesmaschinn” (“reading machine”).
Marco Barnig, a pioneer in language synthesis, revealed to our colleagues from RTL Radio that the foundation for this technology dates back to 2014, when he began developing his own model, “E-Speak.” However, the project remained incomplete until five years ago, when advancements in phoneme technology made it feasible to continue. With the rise of artificial intelligence, the need to convert language into phonemes has diminished, allowing models to be trained directly from text, as Barnig explained.
The voice behind “Liesmaschinn” is Max Kuborn, a well-known media personality in Luxembourg. Kuborn underlined the accessibility benefits of the technology, particularly for individuals who struggle with reading. He also highlighted its potential as a learning tool, helping users familiarise themselves with the sounds and structure of Luxembourgish. According to Kuborn, the applications of this technology are vast and will continue to grow as the tool is further refined and expanded.
ZLS Director Alexandre Ecker acknowledged that developing the tool was not without its challenges. He emphasised that the quality of any language model depends heavily on the data used to train it. In this regard, Ecker praised Kuborn for dedicating “weeks and months” to recording material, as well as a “motivated staff member” who processed, checked, and edited the recordings. These recordings were then paired with written example sentences from the Luxembourg Online Dictionary (LOD). “I think that was the most work-intensive stage,” Ecker noted.
The ZLS highlighted its commitment to preventing misuse of the tool, particularly to ensure that Kuborn’s voice is not exploited for fraudulent activities.
However, the programme remains a work in progress, with ongoing improvements being made. Since no user data is stored, the ZLS relies on feedback to identify and address any issues that arise.
The platform, designed as a public service, is freely accessible to all. Hosted on the website “Sproochmaschinn” (“language machine”), it combines the functionalities of “Schreifmaschinn” and “Liesmaschinn”.
