Not much has been – ahem – said about the Larynx speech synthesiser according to our brief searches online, which returned solely medical results. However, despite the fact that Larynx hasn’t received as many plaudits as the popular Festival, it’s more capable and based on newer studies.
The engine provides a complete text-to-speech solution for nine languages and offers as many as 50 voices. Most of them serve English and German, but there are also variants for French, Italian, Spanish, Russian, Dutch, Swedish and Swahili. Larynx is based on two other open source projects that you can quickly find on Github. These are Gruut for phonemes detection, and Open Neural Network Exchange (ONNX) for machine learning. We won’t go much deeper for the sake of brevity, but it’s enough to know that the speech quality is decent in most languages. It’s not as advanced as in Google Translate, but at the same time it’s good enough to make you stay away from proprietary cloud services. Moreover, Larynx is very easy to set up. In order to give it a try, fetch and then run the following script: raw. githubusercontent.com/rhasspy/larynx/master/ docker/larynx-server. It will download the Larynx Docker image with all its dependencies and immediately run it on your system.