Query:
I’m conducting a quality analysis of Zoom Cloud’s automatic transcription service for code-switched audio (English-Hindi mixing, commonly called “Hinglish”).
Observed Behavior: During transcription of a meeting with mixed English-Hindi speech, I noticed:
- Hindi words are frequently deleted - Words like “matlab”, “thoda sa”, “tha”, “ka” are completely missing from the transcript
- Attempted phonetic transliteration - Some Hindi words are converted to English phonetics:
- “ki” → “key”
- “aapne jo” → “aapneemi bowlanger”
- “sad” → “sadka”
- “hota” → “hotel”
- Entire Hindi phrases dropped - Multi-word Hindi expressions like “aapne jo hindi me bola hai jo ground truth tha” are completely omitted
Questions:
- What is Zoom Cloud’s expected behavior when encountering non-English (specifically Hindi) words in an English-language meeting recording?
- Does it attempt phonetic transliteration?
- Does it skip/ignore Hindi words?
- Can I enable a specific language setting for code-switched content?
- Is there a configuration option to handle code-switched/multilingual audio better?
- Can I set primary + secondary language detection?
- Is there a Hinglish or code-switching mode?
- Language detection: Does Zoom Cloud automatically detect and switch between languages, or does it stick to the meeting’s primary language setting?
Goal: Understand if this is expected behavior or if there’s a configuration to better handle code-switched audio.