term-llm
Docs → Guides → Transcription

Audio

Transcription

Transcribe audio files to text with OpenAI, Mistral Voxtral, Venice, ElevenLabs, a local Whisper server, or whisper.cpp CLI.

On this page

Basic usage

term-llm transcribe meeting.m4a

Supported input extensions include:

  • .ogg
  • .mp3
  • .wav
  • .m4a
  • .flac
  • .mp4
  • .webm

Useful flags

term-llm transcribe interview.mp3 --language en
term-llm transcribe note.m4a --provider openai
term-llm transcribe memo.wav --provider mistral
term-llm transcribe hello.mp3 --provider venice --model nvidia/parakeet-tdt-0.6b-v3
term-llm transcribe hello.mp3 --provider elevenlabs --model scribe_v2
term-llm transcribe call.ogg --provider whisper-cli --porcelain

Key options:

  • --language for a language hint such as en or ja
  • --provider to select the transcription backend
  • --model to override the configured transcription model
  • --timestamps to ask Venice for timestamp metadata before extracting the transcript text
  • --porcelain to output only transcript text

Provider options

term-llm supports several transcription backends:

  • openai
  • mistral (Voxtral)
  • venice
  • elevenlabs
  • local for a local Whisper-compatible server
  • whisper-cli for whisper.cpp

If you omit --provider, term-llm uses the configured transcription provider or falls back to OpenAI.

Venice models

Venice uses POST /api/v1/audio/transcriptions and supports:

  • nvidia/parakeet-tdt-0.6b-v3
  • openai/whisper-large-v3
  • fal-ai/wizper
  • elevenlabs/scribe-v2
  • stt-xai-v1

ElevenLabs models

ElevenLabs uses POST /v1/speech-to-text and supports:

  • scribe_v2
  • scribe_v1

Configuration

transcription:
  provider: venice
  venice:
    api_key: ${VENICE_API_KEY}
    model: nvidia/parakeet-tdt-0.6b-v3
  elevenlabs:
    api_key: ${ELEVENLABS_API_KEY}
    model: scribe_v2

Credential fallback order:

  • Venice: transcription.venice.api_key, VENICE_API_KEY, audio.venice.api_key, image.venice.api_key, or providers.venice.api_key
  • ElevenLabs: transcription.elevenlabs.api_key, ELEVENLABS_API_KEY, XI_API_KEY, audio.elevenlabs.api_key, or providers.elevenlabs.api_key

whisper.cpp CLI mode

For --provider whisper-cli, term-llm looks for a whisper binary in PATH and a model file via:

  • WHISPER_MODEL
  • providers.local_whisper.model in config
  • common default model paths

Example:

export WHISPER_MODEL=~/.local/share/whisper/models/ggml-base.bin
term-llm transcribe note.m4a --provider whisper-cli

When to use it

Use transcription when you want to:

  • turn voice notes into text before summarizing or editing
  • capture meeting audio for later analysis
  • feed transcripts into ask, edit, or your own downstream tooling