Automatic transcription
Transcription done by speech recognition software, without anyone listening to the recording and typing it out.
Automatic transcription uses a speech recognition model to turn audio into text. Compared with manual transcription it’s far quicker and cheaper: an hour-long meeting can be transcribed in well under an hour. The trade-off is mistakes you should review, especially proper names, acronyms and poor-quality audio.
Its quality is usually measured with the word error rate (WER). The model, the language, background noise, people talking over each other and distance from the microphone all play a part.
- Record close to the microphone and with as little noise as possible.
- Add the names, brands and acronyms that will come up to the glossary.
- Review the words the system flags as uncertain.
Echo transcribes with models running on its own servers in the European Union. In our test with 80 Spanish sentences from the public FLEURS dataset it got 97.4% of words right; the method and the limits of the figure are on the accuracy page.
