Voice Cloned Audiobook Generator
Upload a PDF, clone your voice from a short clip, and hear its answers read back in your own voice.
SWE (AI/MLOps)Voice AIRAGLangChainText-to-Speech
A voice-cloned RAG chatbot: upload a PDF, and LangChain plus FAISS build a retrieval index over its text; ask a question and Gemini generates a grounded answer from the retrieved context; the response is then spoken back in a voice cloned from a 3 to 15 second audio sample via AllVoiceLab's MCP (Model Context Protocol) text-to-speech. The whole flow runs through a Streamlit interface, upload a PDF, record or upload a voice sample, ask a question, with voice cloning and speech synthesis running locally after initial setup.
What I built
- Built a RAG pipeline over uploaded PDFs using LangChain and FAISS for vector storage and retrieval.
- Integrated AllVoiceLab's MCP (Model Context Protocol) to clone a user's voice from a 3 to 15 second audio sample and synthesize spoken responses in that cloned voice.
- Used the Gemini API to generate answers grounded in retrieved PDF context rather than the model's own unguided knowledge.
- Built a Streamlit interface covering the full flow: PDF upload, voice sample recording/upload, query input, and cloned-voice playback.
- Kept voice cloning and TTS running locally after initial setup, so spoken responses don't require a round trip to a cloud TTS API per query.
Tech stack
SWE (AI/MLOps)Voice AIRAGLangChainText-to-Speech