Initial situation: There are 6 recorded screencasts (~90 minutes total) from internal incident response and investigation (SIEM) training for the SOC team. The video is in English, there are many terms and abbreviations, a speaker with an accent, there are names of event fields, commands and fragments of configs. Materials need to be released in Russian for Russian-speaking analysts, while maintaining the accuracy of wording and reference to time codes. Technical environment and subject area: cybersecurity - SOC, SIEM, correlation rules, analysis of Windows Event Logs, Sysmon, MITER ATT&CK, basic PowerShell commands, analysis of network indicators. The video contains screen fragments with the console, Kibana/Elastic, Splunk-like interface, tables, JSON fields, regular expressions. What needs to be done is several steps:
1) Extract the speech and prepare an accurate EN transcription with time codes and marks of unclear places - ASR is allowed, but final manual editing is required, especially for terms, numbers, names of techniques and commands.
2) Make a translation EN->RU taking into account domain terminology. The translation should not be literal, but educational and practical - without loss of meaning, but with careful localization of terms where appropriate. Do not translate commands, field names, values, CVEs, MITER techniques, indicators, paths and registry keys, but save them as in the original.
3) Create a glossary (EN term - approved RU version - note) of at least 60 units so that further videos can be translated uniformly.
4) Prepare subtitles in SRT format for each video - with correct segmentation, readability and synchronization. Limitations on subtitles: 1-2 lines, up to 42 characters per line, reading speed no higher than 17 cps, on screen for at least 1.0 sec and no more than 6.0 sec, without overlapping timecodes.
5) Conduct quality control - catch inconsistencies in terms, out of sync, errors in numbers, commands and field names, and issue a short report on controversial areas with suggestions. Specific result: a folder with 6 .srt (RU) files and 6 .txt or .docx files with the final EN transcription and RU translation by paragraph, a single .xlsx glossary, and a .txt QC report with a list of problem fragments by timecodes. Restrictions: original videos and possible internal titles should not be published or used for training models. Do not change the meaning of incident response instructions. Any doubtful places should be noted, and not guessed at. It is important not to break technical artifacts - commands, requests, paths, field names, hashes and IP must match the original. Measurable acceptance criteria:
- 100% coverage of timecodes - everything spoken is reflected in the subtitles, without gaps longer than 2 seconds, except for pauses.
- The discrepancy between the duration of replicas in SRT and audio is no more than 200 ms in key places (commands, indicators, conclusions) and no more than 400 ms on average.
- Errors in numbers, CVEs, paths, commands and field names - 0 based on the results of a random check of 30 random fragments.
- Uniformity of terms: at least 95% matches with the glossary for checking occurrences (controversial terms may be marked in the QC report).
- Compliance with line length and cps restrictions - at least 95% of lines, the remaining cases are justified (for example, a long command) and noted. Professional tools: it’s more convenient to work in Subtitle Edit/Aegisub; Whisper-like solutions are allowed for transcription, but final synchronization and manual proofreading are required.
