Voice Clone TTS

Type any text, hear it in your own voice. Runs fully offline.

---

Setup (first time only)

1. Install system packages:

sudo apt install portaudio19-dev python3-tk espeak-ng -y

2. Install Python 3.10 via pyenv (required on Debian to avoid lzma bug):

curl https://pyenv.run | bash
echo 'export PYENV_ROOT="$HOME/.pyenv"' >> ~/.bashrc
echo '[[ -d $PYENV_ROOT/bin ]] && export PATH="$PYENV_ROOT/bin:$PATH"' >> ~/.bashrc
echo 'eval "$(pyenv init - bash)"' >> ~/.bashrc
source ~/.bashrc
pyenv install 3.10.14
pyenv local 3.10.14

3. Create a virtual environment:

~/.pyenv/versions/3.10.14/bin/python -m venv .venv
source .venv/bin/activate

4. Install Python packages (takes 15–30 min, downloads ~2GB):

pip install --upgrade pip
pip install --no-cache-dir "numpy==1.22.0"
pip install --no-cache-dir --resume-retries 20 TTS sounddevice scipy
pip install --no-cache-dir "transformers==4.40.0"
pip install --no-cache-dir "torch==2.1.0" "torchaudio==2.1.0"

---

Running the app

Every time you want to use it:

source .venv/bin/activate
python voice_clone_tts.py

---

How to use

1. Wait for "Model ready" in the top right (first launch only: downloads ~2GB, takes 5–15 min)
2. Click Start Recording → read the passage below for 30 seconds → Stop Recording
3. Type any text
4. Click Generate & Play

Your voice sample saves as my_voice_sample.wav and is reused automatically on future runs.

---

Best text to record (Rainbow Passage)

"When the sunlight strikes raindrops in the air, they act as a prism and form a rainbow. The rainbow is a division of white light into many beautiful colors. These take the shape of a long round arch, with its path high above, and its two ends apparently beyond the horizon. There is, according to legend, a pot of gold at the end of the rainbow. The shape of a rainbow reminds me of a bridge. Like a bridge, a rainbow is wide in the middle and narrow at its ends."

Read it twice through at your normal pace.

---

Tips

• Record in a quiet room with no background noise
• Speak naturally — don't put on a "reading voice"
• 30 seconds of clean audio is the sweet spot
• Generation takes 10–30 seconds per sentence on CPU