Deepgram ASR (Nova-3)

Transcribe audio files via Deepgram's Nova-3 model. Sub-300ms latency, 50+ languages, $0.0043/min. Free $200 credit on signup.

Transcriba archivos de audio con el modelo Nova-3 de Deepgram. Latencia inferior a 300ms, más de 50 idiomas, $0.0043/min. $200 de crédito gratis al registrarse.

Sending audio to OpenClaw

Currently, audio files can be sent to OpenClaw via Discord or WhatsApp. Send the audio file in a chat message and ask the bot to transcribe it.

Actualmente, los archivos de audio se pueden enviar a OpenClaw a través de Discord o WhatsApp.

> Note: Direct voice recording in the OpenClaw web UI is not yet supported. Use a messaging app to send pre-recorded audio files.

>

> Nota: La grabación de voz directa en la interfaz web de OpenClaw aún no está disponible. Use una aplicación de mensajería para enviar archivos de audio pregrabados.

Quick start

{baseDir}/scripts/transcribe.sh /path/to/audio.m4a

Defaults:

• Model: nova-3
• Output: .txt

Useful flags

{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --out /tmp/transcript.txt
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --language es  # Spanish / Español
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --json --out /tmp/result.json
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --speakers  # speaker diarization / separación de hablantes
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --paragraphs  # smart paragraphs / párrafos inteligentes
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --summarize  # AI summary / resumen con IA
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --detect-language  # auto-detect language / detección automática de idioma

How it works

The script sends audio directly to Deepgram's API (api.deepgram.com) via HTTPS. No third-party services are involved — audio goes only to Deepgram.

1. Reads the local audio file
2. POSTs it to https://api.deepgram.com/v1/listen with your API key
3. Parses the JSON response and extracts transcript text
4. Saves to output file

> Privacy: Audio is sent directly to Deepgram's servers over HTTPS. No data is stored by this skill; Deepgram's data retention policy applies.

Dependencies

• curl — for API calls
• python3 — for JSON response parsing (stdlib only, no pip packages needed)

Credentials

1. Sign up at https://console.deepgram.com
2. Create an API key at https://console.deepgram.com/project/api-keys
3. Free tier: $200 credit on signup

Set the environment variable:

export DEEPGRAM_API_KEY="your_api_key"

Supported languages

Nova-3 supports 45+ languages. Common language codes:

Use --detect-language to auto-detect, or --language to specify.

Full list: https://developers.deepgram.com/docs/models-languages-overview

Supported formats

WAV, MP3, MP4, M4A, OGG, FLAC, WebM, and more.