> ## Documentation Index
> Fetch the complete documentation index at: https://docs.60db.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Speech-to-Text

> STT commands for transcribing audio files

# Speech-to-Text (STT)

## Commands

### Transcribe Audio File

```bash theme={null}
60db stt:transcribe --file audio.wav
```

By default the language is auto-detected. Equivalent to passing `--language auto`.

### Specify Language

```bash theme={null}
60db stt:transcribe --file audio.wav --language hi
```

Pass an ISO 639-1 code from the supported set (see `60db stt:languages`). Supported codes include `en`, `hi`, `bn`, `mr`, `pa`, `gu`, `or`, `as`, `ne`, `ta`, `te`, `kn`, `ml`, `sa`, `ar`, and 25 European languages.

### Auto-detect (explicit)

```bash theme={null}
60db stt:transcribe --file meeting.wav --language auto
```

`--language auto` is treated as "omit" — the CLI strips it before sending so the server runs its language identification across all 39 supported languages. You can also simply omit the `--language` flag entirely for the same behavior.

### Enable Speaker Diarization

```bash theme={null}
60db stt:transcribe --file meeting.wav --diarize true
```

Each response segment will include a `speakers` array with labels like `SPEAKER_00`, `SPEAKER_01`. Adds \~50–150 ms of processing latency.

### Add Context (optional refinement)

```bash theme={null}
60db stt:transcribe --file visit.wav \
  --context "Cricket coaching session. Players: Arjun Mehta, Ishaan Verma. Discussing batting technique."
```

When `--context` is supplied, the server runs a background LLM refinement pass and the returned `text` is polished for proper nouns, filler removal, and punctuation. Omit `--context` to skip refinement.

<Note>
  `--context` takes a **plain string** — the REST `/stt` endpoint shape. The WebSocket `/ws/stt` endpoint accepts a structured `{general, text, terms}` **object** instead; see the [WebSocket STT reference](/api-reference/websocket/stt).
</Note>

### List Available Languages

```bash theme={null}
60db stt:languages
```

Returns the 39-language catalog plus the `auto` entry, sourced from `GET /stt/languages`.

## Options

* `-f, --file <path>` — Audio file path (required; max 25 MB; WAV / MP3 / M4A / OGG / FLAC / WebM)
* `-l, --language <code>` — ISO 639-1 language code (e.g. `en`, `hi`, `ar`). Omit or pass `auto` for auto-detection across the 39 supported languages.
* `--diarize <boolean>` — Enable speaker diarization (default: `false`)
* `--context <text>` — Free-form paragraph describing the session (domain, speakers, jargon). Enables server-side LLM refinement of the response text.

## Examples

### STT — Transcribe Audio

```bash theme={null}
# Auto-detect language (simplest)
60db stt:transcribe --file meeting.wav

# Auto-detect, explicit form
60db stt:transcribe --file meeting.wav --language auto

# Force Hindi — skips language identification for lowest latency
60db stt:transcribe --file recording.wav --language hi

# Multi-speaker meeting with diarization + auto-detect
60db stt:transcribe --file board-call.wav --diarize true

# List supported languages
60db stt:languages
```

## Notes

* **Do not pass unsupported language codes.** The server explicitly rejects `ur`, `ja`, `ko`, `zh`, `th`, `vi`, `id`, `tl`, `sw`, `tr`, `fa`, `he` (and Arabic dialect tags like `ar-eg`) with an `unsupported_language` error. Use `auto` or omit the flag for these.
* **Auto-detect on REST vs WebSocket.** The REST (`stt:transcribe`) form flow accepts `--language auto` as a convenience. The streaming WebSocket form requires `languages: null`, not the literal string `"auto"`. If you write your own WebSocket client, don't forward the string `"auto"` — send `null`.
* A successful response with empty `text` and `warning_codes: ["no_speech_detected"]` means the audio contained no speech (silence / music / noise). This is not an error — do not retry.