In order to illustrate to what extent modern speech recognition is capable of transcribing speech, we have selected a couple of examples.
The first tab below shows the possiblities of a) ASR for Dutch and Italian with high-quality audio and the effect of automatic translation from Dutch/Italian into English. The translation was done with the GoogleTranslate-service in SubtitleEdit. Currently (after 2019) the translations of DeepL are outperforming Google Translate (to our opinion).
The second tab shows the result English ASR for high quality recordings (quiet) environment, good microphones, no background noise) of two lectures by non-natives (a Dutch and an Italian speaker). The third tab shows the outcome of the English ASR when applied on low-quality recordings. The speakers are an English General and a Polish survivor of the Sobibor-concentration camp who lives for already a long time in the USA.
All the recognitions of the English-spoken material were done with the WebASR of the University of Sheffield. The Dutch-spoken recording with the UTwente-Radboud ASR-engine and the Italian-spoken material with the (offline) ASR-engine of Piero Cosi.
- Dr. Laura Hollink (native speaker of Dutch, Amsterdam)
- Manuela Ruisi (native speaker of Italian, Rome)
These lectures were recorded at the CLARIN-PLUS Workshop "Working with Parliamentary Records" (2017).
- General Julian H. A. Thompson (native speaker of British-English, London)
- Regina Zielinski-Feldman (native speaker of Polish, USA)