transcribe

The transcribe package provides an R interface to audio transcription using Whisper, with optional post‑processing via Ollama. It includes a command‑line interface (CLI) and a Plumber API to create a web interface.

Prerequisites

• Whisper:
  Make sure you have OpenAI Whisper installed and configured. Follow the official documentation for installation instructions.

• Ollama:
  Ensure that Ollama is installed and running. Consult the Ollama documentation for setup details and configuration.

• Other Dependencies:
  This package uses:

  • processx to wrap the yt-dlp command for downloading audio files.
  • reticulate to call Python’s Whisper implementation.
  • ellmer for prompt-based post‑processing of raw Whisper transcripts.
    Please refer to each package’s documentation for further details.

Overview

The package workflow is as follows:

• Audio Downloading:
  When given a remote URL, processx is used to call yt-dlp, downloading the audio file quickly and robustly.

• Transcription via Whisper:
  Python’s Whisper is called via reticulate, providing state‑of‑the‑art transcription directly from R.

• Post‑processing with Ollama and ellmer:
  The raw transcript from Whisper is optionally sent to Ollama via ellmer using a prompt (e.g., “Reformat the transcript into clear, well‑punctuated paragraphs…”) to produce a cleaned, readable transcript.

• Interfaces:
  Use the CLI for batch processing, or launch the Plumber API to access a web interface.

Installation

# install.packages("remotes") # if not already installed
remotes::install_github("brancengregory/transcribe")

Usage

Basic Example

library(transcribe)

# Transcribe a local audio file.
transcript <- transcribe_audio(
  input_path = "path/to/audio.wav",
  language = "en",
  whisper_model_name = "large-v3-turbo",
  processed = TRUE,
  ollama_model = "llama3.2"
)
cat(transcript)

CLI Usage

To use the command‑line interface, run the following command from your terminal:

Rscript inst/scripts/main_cli.R -i "path/to/audio.wav" -l en -m large-v3-turbo -p TRUE -M llama3.2 -o "transcribe.txt"

This command processes the specified audio file and saves the transcript to transcribe.txt.

Plumber API

You can serve a web interface via plumber. For example:

library(plumber)
plumber::plumb("inst/plumber/api.R")$run(port = 7608)

Then open your browser at http://127.0.0.1:7608 to use the transcription interface.

Vignettes

For a detailed introduction, see the vignette “intro”:

vignette("intro", package = "transcribe")

Contributing

1. Fork the repository on GitHub.
2. Create a new branch for your changes.
3. Submit a pull request describing your proposed changes.

License

MIT © 2025 Your Name