Speech Recognition System for Slovenian Language

This public procurement contract involves the purchase, customization, and installation of a speech recognition system for the Slovenian language. The details of the procurement are outlined in section 3.3 Technical Requirements, where the contracting authority specifies the technical, functional, and process requirements based on the specific needs for subtitling audiovisual content at the public institution Radiotelevizija Slovenija. The requirements include the purchase, installation, commissioning, and customization of server software for Slovenian speech recognition. The contracting authority also requires the development, installation, and commissioning of client software for delayed transcription and for monitoring and managing the system's performance. The goal of this procurement is to establish a modern, neural network-based end-to-end system with the latest architectures (e.g., transformer or similar) for Slovenian speech recognition, which will provide core technical functionality for the processes involved in preparing subtitles. Due to the specific requirements of the field and compliance with legal quality standards, the supplier must offer a system that can be tailored to linguistic and technical requirements and specialized using the contracting authority's training data for at least three specific areas of application: informative, cultural, and sports content. The contracting authority will prepare a training corpus with at least 100 hours of audio recordings with manually prepared and verified verbatim reference transcripts for each area. The server software must be designed modularly, with clearly defined components to allow for gradual expansion of functionality and performance, as well as easy upgrades to new versions of models and software. All recognition processes will take place on the contracting authority's infrastructure in its data center, unless explicitly stated otherwise in this procurement documentation. The supplier must include a technical specification of the offered server software and all other evidence, certificates, and documentation required by the contracting authority in the technical specifications of the equipment. The subject of this public procurement includes: • Purchase of a server license, which means purchasing a permanent license for all server components of the software necessary for continuous real-time recognition of input speech in audiovisual content and transcription of speech on demand. • Installation of the server and client application on the contracting authority's hardware and customization, which includes the installation of the server along with initial and later specialized models and the client application on the contracting authority's hardware in its data center. It also includes the installation of system software necessary for establishing monitoring modules to track the system metrics of the recognizer. • Development of a project for implementation (PZI) and development of client software for using the speech recognizer in the contracting authority's environment according to the PZI: this includes creating a development plan (PZI) for a web application, developing a custom web application for delayed transcription and editing of transcripts, including its installation; • Customization of the Slovenian speech recognition model, which includes making adjustments or specialization of the recognition model to specific domains based on training data provided by the contracting authority.