透過Local LLM、RAG、TTS、STT等技術所完成的專案
This project was made by Local LLM、RAG、TTS and STT
Note
此項專案利用Meta 提出的llama_cpp套件引用 Hugging face 中的 pre-trained model(Mistral 7B)
This project utilizes the llama_cpp library proposed by Meta to reference the pre-trained model (Mistral 7B) from Hugging Face.
建議安裝CUDA版本
recommend running this project using the CUDA version of PyTorch.
權重/weight
mistral-7b-instruct-v0.1.Q3_K_M.gguf
將下載的權重放置於 Place your weight under the LLM_weight
TTS-System └──backEnd └──LLM_weight └── .gguf file
預覽/preview
Note
使用者可以透過文字輸入或語音輸入與系統互動,系統會以語音加文字的方式進行回饋
Users can interact with the system by speaking or typing, and will receive feedback through voice and text.
使用者可以透過上傳TXT檔的方式使系統回答得更加準確
Users can get hight qulity response by upload txt file.
Sentence Transformers:sentence_transformers
這項專案使用 sentence_transformers 作為詞向量轉換的套件,並利用餘弦相似度算出前 K 個相關的資訊提供給prompt
This project uses the sentence_transformers library for word embeddings and employs cosine similarity to calculate the top K relevant pieces of information to provide for the prompt.
Note
💡 為了確保語音合成相關功能可以使用,請先Clone VALL-E-X 中的專案自行測試
💡 需確保能夠正常啟用React.JS相關專案
💡 需安裝FFMPEG相關檔案
Clone VALL-E-X and make sure you can run this project!!!
make sure React.JS can be use!!!
need to download FFMPEG
cd TTS-Systeam
python -m venv env
Note
激活虛擬環境/Activate the virtual environment
Unix/macOS
source env/bin/activate
Windows
.\env\Scripts\activate
pip install -r requirements.txt
這部份需要兩個 Terminal,分別用於前端與後端
This part need tow Terminal,one for backend, the other for front
cd backEnd
python main.py
cd web-tts
npm start