formadi / piper

A fast, local neural text to speech system

A fast, local neural text to speech system that sounds great and is optimized for the Raspberry Pi 4. Piper is used in a variety of projects.

echo 'Welcome to the world of speech synthesis!' | \
  ./piper --model en-us-blizzard_lessac-medium.onnx --output_file welcome.wav

Listen to voice samples and check out a video tutorial by Thorsten Müller

Voices are trained with VITS and exported to the onnxruntime.

Voices

Our goal is to support Home Assistant and the Year of Voice.

Download voices for the supported languages:

• Catalan (ca)
• Danish (da)
• German (de)
• British English (en-gb)
• U.S. English (en-us)
• Spanish (es)
• Finnish (fi)
• French (fr)
• Greek (el-gr)
• Icelandic (is)
• Italian (it)
• Kazakh (kk)
• Nepali (ne)
• Dutch (nl)
• Norwegian (no)
• Polish (pl)
• Brazilian Portuguese (pt-br)
• Russian (ru)
• Swedish (sv-se)
• Ukrainian (uk)
• Vietnamese (vi)
• Chinese (zh-cn)

Installation

Download a release:

• amd64 (64-bit desktop Linux)
• arm64 (64-bit Raspberry Pi 4)
• armv7 (32-bit Raspberry Pi 3/4)

If you want to build from source, see the Makefile and C++ source. You must download and extract piper-phonemize to lib/Linux-$(uname -m)/piper_phonemize before building. For example, lib/Linux-x86_64/piper_phonemize/lib/libpiper_phonemize.so should exist for AMD/Intel machines (as well as everything else from libpiper_phonemize-amd64.tar.gz).

Usage

1. Download a voice and extract the .onnx and .onnx.json files
2. Run the piper binary with text on standard input, --model /path/to/your-voice.onnx, and --output_file output.wav

For example:

echo 'Welcome to the world of speech synthesis!' | \
  ./piper --model en-us-lessac-medium.onnx --output_file welcome.wav

For multi-speaker models, use --speaker <number> to change speakers (default: 0).

See piper --help for more options.

People using Piper

Piper has been used in the following projects/papers:

• Home Assistant
• Rhasspy 3
• NVDA - NonVisual Desktop Access
• Image Captioning for the Visually Impaired and Blind: A Recipe for Low-Resource Languages
• Open Voice Operating System

Training

See src/python

Pretrained checkpoints are available on Hugging Face

Start by installing system dependencies:

sudo apt-get install python3-dev

Then create a virtual environment:

cd piper/src/python
python3 -m venv .venv
source .venv/bin/activate
pip3 install --upgrade pip
pip3 install --upgrade wheel setuptools
pip3 install -r requirements.txt

Run the build_monotonic_align.sh script in the src/python directory to build the extension.

Ensure you have espeak-ng installed (sudo apt-get install espeak-ng).

Next, preprocess your dataset:

python3 -m piper_train.preprocess \
  --language en-us \
  --input-dir /path/to/ljspeech/ \
  --output-dir /path/to/training_dir/ \
  --dataset-format ljspeech \
  --sample-rate 22050

Datasets must either be in the LJSpeech format (with only id/text columns or id/speaker/text) or from Mimic Recording Studio (--dataset-format mycroft).

Finally, you can train:

python3 -m piper_train \
    --dataset-dir /path/to/training_dir/ \
    --accelerator 'gpu' \
    --devices 1 \
    --batch-size 32 \
    --validation-split 0.05 \
    --num-test-examples 5 \
    --max_epochs 10000 \
    --precision 32

Training uses PyTorch Lightning. Run tensorboard --logdir /path/to/training_dir/lightning_logs to monitor. See python3 -m piper_train --help for many additional options.

It is highly recommended to train with the following Dockerfile:

FROM nvcr.io/nvidia/pytorch:22.03-py3

RUN pip3 install \
    'pytorch-lightning'

ENV NUMBA_CACHE_DIR=.numba_cache

See the various infer_* and export_* scripts in src/python/piper_train to test and export your voice from the checkpoint in lightning_logs. The dataset.jsonl file in your training directory can be used with python3 -m piper_train.infer for quick testing:

head -n5 /path/to/training_dir/dataset.jsonl | \
  python3 -m piper_train.infer \
    --checkpoint lightning_logs/path/to/checkpoint.ckpt \
    --sample-rate 22050 \
    --output-dir wavs

Running in Python

See src/python_run

Run scripts/setup.sh to create a virtual environment and install the requirements. Then run:

echo 'Welcome to the world of speech synthesis!' | scripts/piper \
  --model /path/to/voice.onnx \
  --output_file welcome.wav

If you'd like to use a GPU, install the onnxruntime-gpu package:

.venv/bin/pip3 install onnxruntime-gpu

and then run scripts/piper with the --cuda argument. You will need to have a functioning CUDA environment, such as what's available in NVIDIA's PyTorch containers.