Yarp device for Whisper

This repository contains the yarp-device-speechTranscription-whisper plugin.

🚧 This repository is currently work in progress. 🚧 🚧 The software contained is this repository is currently under testing. 🚧 🚧 APIs may change without any warning. 🚧

Documentation

Documentation of the nws/nwc devices is provided in the official Yarp documentation page:

Documentation of the interface API is provided in the official Yarp documentation page:

Documentation of the audio in Yarp. https://yarp.it/latest/group__AudioDoc.html

Installation

Step1: Build whisper.cpp library

# Clone whisper.cpp repository, choose an install dir, make it and install it to the chosen install dir.
# Please note that whisper.cpp has several build options. Some of them e.g. the use of the GPU may affect
# performances significantly. The commands reported below refers only to default configuration.
# Please check the documentation on the official page github page.
# ${ROBOT_CODE} is the root directory of your choice.
# ~/my_whispercpp_installation_dir is a directory of your choice (where the whispercpp library will be installed)

 cd ${ROBOT_CODE}
 git clone https://github.com/ggerganov/whisper.cpp -b v1.6.2 whispercpp 
 cd whispercpp
 mkdir build
 cd build
 cmake -GNinja -DBUILD_SHARED_LIBS:BOOL=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON -DCMAKE_INSTALL_PREFIX=~/my_whispercpp_installation_dir ..
 cmake --build .
 cmake --install .

Step 2: Build the yarp device

 cd ${ROBOT_CODE}
 git clone https://github.com/robotology/yarp-device-speechTranscription-whisper
 cd yarp-device-speechTranscription-whisper
 mkdir build
 cd build
 cmake -GNinja -DWHISPER_ROOT=~/my_whispercpp_installation_dir ..
 cmake --build .

Step 3: Install the model(s)

# Here is a list (not complete) of possible whisper models: tiny.en, tiny, base.en, base, small.en, small, medium.en, medium, large-v1, large
  wget -P ${ROBOT_CODE}/yarp-device-speechTranscription-whisper/build/share/WhisperTranscribe/contexts/whisperTranscribe_demo/ggml-base.en.bin https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin

Usage

yarp-device-speechTranscription-whisper is a yarp device and it cannot be executed as a standalone module. It requires to be attached to an nws device, e.g. SpeechTranscription_nws_yarp and requires a yarprobotinteface configuration file to be correctly instantiated.

Check the demo examples provided in: https://github.com/robotology/yarp-device-speechTranscription-whisper/tree/master/src/devices/whisperSpeechTranscription/demos demo_audio_from_file.xml demonstrate how to transcribe audio from a recorded file. demo_audio_from_mic.xml demonstrate how to transcribe audio from a microphone. The audio is also optionally recorded to a file. The transcribed text is provided on the port /speechTranscription_nws/text:o

CI Status

🚧 This repository is currently work in progress. 🚧

License

🚧 This repository is currently work in progress. 🚧

Maintainers

This repository is maintained by:

	@randaz81