This project aims to fully digitize and automate the transcription of spoken dialogues. We will implement a software system, Interscriber, that takes an audio recording as input and creates text using algorithms for Speech-to-Text and Speaker Diarization. The text is further processed and corrected. Finally, Interscriber applies semantic analyses such as topic modeling, sentiment analysis and summarization to extract key insights, which can serve as the basis to write news articles, communications, or meeting minutes. All services run on Swiss servers since interviews may contain sensitive data. The demand for automatic solutions for reliable transcriptions is shaped by the huge effort for manual transcriptions, combined with recent advances in research on speech-to-text that make auto-transcriptions feasible.

Target users of Interscriber are everyone who creates transcriptions and their summaries manually, e.g. journalists, secretaries, social scientists, or bank consultant. Interscriber targets the DACH region. Thus, it will support German, English, and Swiss German (the latter will be developed in a separate project). By providing a market-ready tool like Interscriber, writing undergoes digital transformation where users: i) make use of ML for generating reliable transcriptions, ii) reduce workload for post-editing and iii) base their
intellectual work on automatically extracted insights. This allows them to dedicate more time to meaningful tasks. Interscriber’s core technological innovation stems from enhancing existing transcription methods to increase their quality on interviews and dialogues, which adds the challenges of low audio quality, overlapping and spontaneous speech.

SpinningBytes is a Swiss startup founded in 2015 that develops solutions for Natural Language Processing (NLP) based on machine learning (ML) algorithms. With this project, it is expanding its services by providing a new product “Interscriber” for automatic interview transcription.