NeuTTS Android (Termux Unofficial)

This is an unofficial, community-maintained repository for running Neuphonic's NeuTTS (Text-to-Speech) and Llama.cpp (LLM) locally on Android devices via Termux.

This build includes custom workarounds for hardware-accelerated inference on Snapdragon-based devices, enabling the use of the NPU (Hexagon DSP) and Adreno GPU.

Features

• NeuTTS Voice Cloning: Local inference for voice synthesis.
• Gradio Web Interface: A unified dashboard to handle voice synthesis and LLM chat.
• Hardware Acceleration: Native NPU (Hexagon) and GPU offloading support for Llama models.
• Termux Optimized: Fixes for common Python dependency issues (librosa, llvmlite, sentencepiece, soxr).

Prerequisites

1. Install Termux.
2. Run pkg upgrade.
3. Install necessary build tools: pkg install clang python llvm cmake git.

Quick Start Guide

1. Installation

Clone the repository:

git clone https://github.com/americanumberoneaokgoodtimes/neutts-android-termux-unofficial
cd neutts-android-termux-unofficial

Install dependencies:

pip install -r requirements.txt

2. Configuration

We have included a detailed setup guide for specific environment issues: Read the full Termux Setup Guide

3. Launch the Control Center

python3 app.py

Open http://0.0.0.0:7860 in your Android browser to access the web interface.

---

Hardware Acceleration (Snapdragon 8 Gen 5 / Elite)

This repository leverages the Qualcomm NPU (Hexagon DSP). Ensure your binary is pointed at the correct HTP drivers (as documented in the setup guide) to offload model layers.

Known Workarounds Applied

• Librosa Bypass: Removed in favor of torchaudio to avoid numba circular import panics.
• llvmlite: Uses the pre-compiled pkg install python-llvmlite to avoid LLVM 21 API mismatch issues.
• soxr: Compiled manually from source to handle Android's specific linker requirements for Python 3.13.

---

Disclaimer: This repository is a community-maintained unofficial port and is not affiliated with Neuphonic.