earshot

Shiny app for speech-to-text transcription using stt.api.

Installation

remotes::install_github("cornball-ai/whisper")
remotes::install_github("cornball-ai/stt.api")
remotes::install_github("cornball-ai/earshot")

Usage

RStudio: Open the project and click "Run App" (uses app.R)

From R:

library(earshot)
run_app()

Opens at http://localhost:7802 by default.

Features

• Record from microphone (requires browser - see note below)
• Upload audio files (.wav, .mp3, .m4a, .ogg, .flac, .webm)
• Model selection (tiny, base, small, medium, large, or whisper-1 for OpenAI)
• Optional language hint for improved accuracy
• Optional prompt for names, acronyms, or domain-specific terms
• View transcription text, segments with timestamps, or raw API response

Microphone Recording

To use microphone recording, you must open the app in a web browser (not RStudio's viewer pane). Click "Open in Browser" in the Shiny viewer tab, or navigate directly to the URL (e.g., http://localhost:7802).

Microphone access requires:

• A secure context (HTTPS or localhost)
• Browser permission to access the microphone

Backends

earshot uses stt.api which supports multiple transcription backends. In auto mode, backends are tried in this order:

1. whisper (native R torch) - Fastest, runs locally, no API key needed
2. api (OpenAI or compatible) - Requires API endpoint and key
3. audio.whisper - Fallback using the audio.whisper package

Native whisper (recommended)

Install the whisper package for local transcription with no API dependencies:

remotes::install_github("cornball-ai/whisper")

Models are downloaded automatically on first use.

OpenAI API

To use OpenAI's API, set your API key in ~/.Renviron:

OPENAI_API_KEY=sk-...

Then configure stt.api:

stt.api::set_stt_base("https://api.openai.com")
stt.api::set_stt_key(Sys.getenv("OPENAI_API_KEY"))

Local whisper server

For a local OpenAI-compatible server (e.g., whisper container):

stt.api::set_stt_base("http://localhost:8200")

audio.whisper

Install from the bnosac drat repository or GitHub:

# From drat
install.packages("audio.whisper", repos = "https://bnosac.github.io/drat")

# From GitHub
remotes::install_github("bnosac/audio.whisper")

License

MIT