Cohere launches an open source voice model specifically for transcription

Relatively light at just 2 billion parameters, the model is meant for use with consumergrade GPUs for those who want to selfhost it. It currently supports 14 languages.