Use openai-whisper or faster-whisper for transcribing audio
The _transcribe nodes transcribe each incomming audio_common_msgs/AudioData. Use these with a node that records audio into a single frame e.g. the silero_vad node.
package openai-whisper for whsiper
package faster-whisper for faster-whisper
To use this in a ros space make sure the packages are available e.g. by installing them into the default ROS python path
export PYTHONUSERBASE=/path/to/workspace/{devel or install}
# we link all possible python paths together so we can simply source setup.bash e.g.
cd $PYTHONUSERBASE/lib
mkdir -p python3/dist-packages
ln -s python3 python3.8
cd python3
ln -s dist-packages site-packages
# install openai-whisper requirement torch for your cuda version (cuda 11.4 -> cu114)
pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu114
pip3 install openai-whisper faster-whisper huggingface-hub==0.11.0
# packages are now useable with default setup script
source /path/to/workspace/devel/setup.bash