SpeechTech Archive technology is a system used for creating and searching in audiovisual archives.
The core of SpeechTech archive technology is automatic speech-to-text transcription and its purpose is to grant the users the ability to quickly find objects (words or phrases) in vast archives of audio or video tracks. SpeechTech Archive allows the user to find not only known words, but also words that were not in the automatic speech recognition system vocabulary at the moment of indexing. This makes it possible to search and find various geographic names as well as names of people or products. The database index is quickly filtered and the result is usually available in a matter of seconds.
First, the audio track of the record is analyzed using methods of automatic speech recognition. Several different hypotheses about the content of each utterance are saved in the database index at the same time. All results (words) are indexed and saved into the database. Afterwards, in the search phase, all of these hypotheses are assessed and displayed as found results along with a link to the corresponding location in the audio archive where it is possible to watch or listen to the record immediately.
- Search based on words or sounds
- Possibility to search for different forms of known words
- Ability to index not only audio tracks but also video tracks that include audio
- Very fast search through the archive (e.g. 1000 hours in a matter of seconds)
- Multiplatform modular architecture – supports client-made modifications
- Can be implemented on a cluster to allow indexing and searching in vast archives
- Client-server architecture
- Accessible using a web or API interface