The hardware and bandwidth for this mirror is donated by dogado GmbH, the Webhosting and Full Service-Cloud Provider. Check out our Wordpress Tutorial.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]dogado.de.
Native R torch implementation of OpenAI Whisper for speech-to-text transcription.
# Install dependencies
install.packages(c("torch", "hfhub", "safetensors", "av", "jsonlite"))
# Install whisper from GitHub
remotes::install_github("cornball-ai/whisper")library(whisper)
# Transcribe the bundled JFK "Ask not" speech (prompts to download model on first use)
jfk <- system.file("audio", "jfk.mp3", package = "whisper")
result <- transcribe(jfk)
result$text
#> "Ask not what your country can do for you, ask what you can do for your country."On first use, you’ll be prompted to download the model:
Download 'tiny' model (~151 MB) from HuggingFace? (Yes/no/cancel)# Download a model explicitly
download_whisper_model("tiny")
# List available models
list_whisper_models()
#> [1] "tiny" "base" "small" "medium" "large-v3"
# Check which models are downloaded
list_downloaded_models()
# Check if a specific model exists locally
model_exists("tiny")# Basic transcription
result <- transcribe("audio.wav")
print(result$text)
# Specify model size
result <- transcribe("audio.wav", model = "small")
# Force CPU (useful if CUDA has issues)
result <- transcribe("audio.wav", device = "cpu")
# Non-English audio (specify language for better accuracy)
allende <- system.file("audio", "allende.mp3", package = "whisper")
result <- transcribe(allende, language = "es")
# Translate to English (quality is model-dependent; larger models work better)
result <- transcribe(allende, task = "translate", language = "es", model = "small")| Model | Parameters | Size | English WER |
|---|---|---|---|
| tiny | 39M | 151 MB | ~9% |
| base | 74M | 290 MB | ~7% |
| small | 244M | 967 MB | ~5% |
| medium | 769M | 3.0 GB | ~4% |
| large-v3 | 1550M | 6.2 GB | ~3% |
Models are downloaded from HuggingFace and cached in
~/.cache/huggingface/ unless otherwise specified.
MIT
These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.
Health stats visible at Monitor.