A-Hackers-AI-Voice-Assistant
A-Hackers-AI-Voice-Assistant copied to clipboard
LearnedVector
installing CTCDECODE
Any solution on installing missing dependencies?
You need to make it from source. Follow the instructions under the ctcdecode readme
Also. Get wsl2. Most of the code in this project does not run in default windows
i followed but it came out like this
@daniele777 do you think that it is possible to use linux? Because wsl2 is a linux virtual machine.
Yes with Linux run ok ! What video card do you have ? I used deepin distribuition he have aleady have python 3.7 install.. Just install sudo apt-get python3-pip
@daniele777 I have an rtx3080:)
Good because i have an amd video card and can't use for train neural network now i see in pytorch there is an amd support But code fault with ROCm pytorch
With CUDA and cpu only can't go up to 2 epoch i use this Command
python3 train.py --sample_rate 8000 --epochs 2 --batch_size 32 --eval_batch_size 32 --lr 0.0001 --model_name ./new_wakeword_v0 --save_checkpoint_path neuralnet/checkpoints --train_data_json data/json/train.json --test_data_json data/json/test.json --num_workers 4 --hidden_size 64
Model was created but wakeword not recognized
@daniele777 I would try to reduce the batch size or work with steps instead of epochs
@daniele777 I don't know but if u work with steps you have to change the code I thing
@daniele777 do you know if Ubuntu and Linux is the same? Because I am about to format my machine to an Ubuntu machine.
I don't know for Ubuntu i used ctcdecode in Kali Linux and deepin distribuitions without problem
@CrackCatZ Yesterday i train model but i have AMD video card,but have different Command on train.py ... and with only CPU can't train well thinking ...sample rate in train have same value of rate of WAV wakeword stored ? If you have time we can share our knolegment about, direcly on telegram or another like discord it's not simple to meet Person like vocal assistant :)