Output modes
The same recording can be written three ways. Pick the one your readers expect — it is a per-request switch, so different products can use different modes.
The three modes
| Mode | Writes | Good for |
|---|---|---|
native | Everything in the language's own script, including English words people used. | Records, search in the native script, formal documents. Default. |
codemix | English words in English, the rest in the language's script; numbers as digits. | Most product UIs, chat, notes, analytics — the way people in India actually type. |
verbatim | Every word as spoken, including fillers, false starts and repeats. | Research, legal and compliance review, speech analytics. |
The same sentence in Hindi:
कल की मीटिंग का टाइम बदलकर चार बजे कर दीजिएकल की meeting का time बदलकर 4 बजे कर दीजिएअं कल की मीटिंग का टाइम टाइम बदलकर चार बजे कर दीजिएChoosing a mode
- Showing text to people who switch between languages? Start with
codemix. - Feeding a system that expects one script per language (search indexes, language-specific tools)? Use
native. - Need to know exactly what was said, hesitations included? Use
verbatim.
Every one of the 22 Indian languages supports all three modes. English always uses native.
Setting the mode
JSON
{ "audio": { "upload_id": "upl_…" }, "language": "ta-IN", "mode": "codemix" }The mode used is echoed in the result's mode field.