Turn a local text file into speech

Install Piper and get a voice

Piper reads the text you give it and produces speech locally. In a Python 3 virtual environment, install the version below. These commands were tested with Python 3.13 on Apple silicon. Check the linked package for a build that matches your platform.

python -m pip install "piper-tts==1.8.0"
mkdir voices
python -m piper.download_voices en_US-lessac-medium --download-dir voices

The download supplies an .onnx model and its matching .onnx.json settings. If the download command fails, save both files from the voice’s official file listing into the voices folder. Consult the voice’s model card for language, speaker, and license information.

Generate a recording

Save this text file in your working directory. Run the command below after both voice files are available.

The reading room closes at six p.m. on Saturdays. Ask at the front desk for help.

python -m piper --model voices/en_US-lessac-medium.onnx --input-file announcement.txt --output-file announcement.wav

Compare two speaking rates

python -m piper --model voices/en_US-lessac-medium.onnx --input-file announcement.txt --output-file fast.wav --length-scale 0.8 --noise-scale 0 --noise-w-scale 0
python -m piper --model voices/en_US-lessac-medium.onnx --input-file announcement.txt --output-file slow.wav --length-scale 1.3 --noise-scale 0 --noise-w-scale 0

Play both files and compare durations. Check whether the slower version is easier to understand. Try writing a time in digits and then in words. Listen for the difference before choosing wording for an announcement.