Transcribing, Translating and Generating Text
Witness interviews which are typically documented as video or audio recording, and other digital or digitized sources like newscasts or movies are common sources for historical, social, medial and other studies. For analyzing or citing such sources, it is usually neccessary to transcribe at least parts of the spoken content. The process of transcribing audio or video recordings takes at least 6 times as long as the speaking time, if you have to right equipment.
If you need to transcribe a recording, you can safe a lot of time by using YouTube’s captioning system and downloading the transcript.
How to transcribe your recording and download the transcript:
- Upload your file to YouTube and hit the transcription button.
- Copy the video URL (of your or any other video published on YouTube) and paste it in a YouTube srt downloader like DownSub. Download the srt-file.
- Use an editing tool like SubTitleEdit to clean the text and get a txt-file which you can edit in Word, for instance:
“File” > “Open” → “Export” > “Plain Text” (adjust settings)
The result will not be highly accurate and needs revision, but this process is much more efficient than starting from the scratch.
Since AI-based tools keep learning, the output is not static and may change over time. The output you get from such translation tools depends on the training it received from its users. Thus, the quality of the translation may vary over time and can differ from tool to tool. At some point you may get more accurate results for certain phrases with one tool, weeks later you might get better results for the same phrase with another tool. Also, be aware that there are people who deliberately train AI incorrectly. Do not use AI-based translations without examining and, if neccessary, correcting the results.
Both Google Translate and DeepL offer the option to type or paste in text or upload documents for translation (Google: doc, docx, odf, pdf, ppt, pptx, ps, rtf, txt, xls, xlsx; DeepL: docx, pptx).
Compare the results from DeepL and Google Translate
Talk to Transformer generates text which completes your snippets. Just “type something and a neural network will guess what comes next”…
While the quality of these text fragments varies highly in terms of content and viability, they are grammatically correct and internally consistent. It is a fun gadget, but critics have raised concerns regarding the online distribution of AI-produced content which readers cannot distinguish from human produced texts, and its implications for the reliability and dangers of information brokerage.