Let's build a conversational engine so we can talk to our computers! Demo with audio
Is this project useful to you? Give me a ⬆money upvote!⬆
Right now, we have been testing this on linux + cuda. The project is still at an early stage, and requires a lot of elbow grease to get running. We'll keep on making it better as time goes on!
Wed Jun 21 2023
The intended audience for this project at the current state is people who are comfortable with hacking things together.
chmod 775 build.sh
./build.sh
If you would like to install piper automatically: (this downloads the piper binaries and the default TTS model)
source install_piper.sh true $([ -n "$BASH" ] && echo 1 || echo 2)
WARNING: The bash script will move the existing config.json
file to config.json.bkp
and create a new one instead.
piper
from anywhere in your system should work.npm install
git submodule init && git submodule update --recursive
npm install
in whisper.cpp/examples/addon.node
cd whisper.cpp && make
cd llama.cpp && make
npx cmake-js compile --CDWHISPER_CUBLAS="ON" -T whisper-addon -B Release
mv build/Release/* ../bindings/whisper/
config.json
to point to the models you downloadedconfig.json
to point to record_audio.sh
to listen from mic or sample_audio.sh
for bundled audio examplesrecord_audio.sh
is selected, make sure sox
package is install in your system. You can install it apt install sox libsox-fmt-all
./llama.cpp/build_server/bin/server -m models/llama/nous-hermes-13b.ggmlv3.q4_K_S.bin -c 2048
npm run start
A graphviz file talk.dot
will be created when you press ctrl-C.
You can view the graph by running npm run graph
, which will plot an svg and open it.
Please do
vim ./${llama/whisper}/examples/addon.node/addon.cpp