Model adaptation

We offer a model customization service to reach this goal in case the generic model is not well suited to your data. Depending on the available data, we can adapt the whole model, or only part of the model such as the pronunciation model, the language model, and the acoustic model. Therefore we can provide accurate tailored solutions that cannot be matched with general-purpose open-source solutions. We also offer solutions optimized for low footprint and low power consumption when needed.

Model development

We also offer a service to build models on demand in order to get the best possible results on your data. Each application needs a specific development plan which is taking into account the availability of linguistic ressources and the specific constraints linked to the usage of these ressources. Our work always starts with a careful study of your needs and of the data to be processed. Then we may need to create or collect some specific linguistic ressources. The rest of the process is the model optimization to maximise the model accuracy on your data. It should be noted that we have demonstrated our ability to get the best of the available language ressources using our technology in our participations to many international challenges which we won.

Security issues

All our AI models are developed using our own data or are adapted from existing models that we have built in-house using that same data. This approach eliminates the risks associated with publicly available or outsourced AI models, which may contain biases, hidden vulnerabilities, or even intentional backdoors (e.g., through data poisoning). For critical operations, we use a glass-box approach, where the sources of knowledge are broken down and clearly defined. This allows for finer control over the models and helps mitigate the risk of learning from corrupted data.