AI text-to-speech has improved enormously, and you no longer need a paid cloud service to get natural-sounding voices. Two free, open-source projects, Kokoro and Piper, run entirely on your own PC, so your text never leaves your machine and there are no usage limits. Both offer British English voices.
Forge PC hasn't tested these tools. The steps and details below come from each project's official GitHub and Hugging Face pages; commands can change between versions, so check the project pages if something doesn't work.
Kokoro or Piper?
| Kokoro | Piper |
|---|
What it is | Open-weight AI voice model with 82 million parameters | Fast local neural text-to-speech engine |
Licence | Apache 2.0 (model weights) | GPL-3.0 (voice licences vary) |
Install | pip install kokoro plus espeak-ng
| pip install piper-tts
|
British English | Yes (several voices) | Yes (en_GB voices) |
Best for | Highest quality from a small model | Speed and lightweight use, including smart home projects |
Kokoro's developers say it delivers quality comparable to larger models while being significantly faster and more cost-efficient. Piper is used in projects including Home Assistant and the NVDA screen reader.
What you need
A Windows 11 PC (both also run on Linux and macOS).
Python 3 installed, with pip. If you're new to Python on Windows, a separate virtual environment for each tool avoids conflicts.
No powerful graphics card is required to try them, though both can use a GPU.
Option 1: Install and use Kokoro
Step 1: Install espeak-ng
Kokoro uses espeak-ng as a fallback for words it doesn't know and for some languages. On Windows, the Kokoro page says to:
Go to the espeak-ng releases page on GitHub.
Download the appropriate .msi installer for your system.
Run the installer.
Step 2: Install Kokoro
Open a terminal and run:
pip install "kokoro>=0.9.4" soundfile
The quotation marks stop the Windows command prompt treating > as a redirect; Kokoro's own example omits them because it runs in a notebook.
Step 3: Generate speech
Create a file called speak.py with this code, based on the example in Kokoro's README:
from kokoro import KPipeline
import soundfile as sf
pipeline = KPipeline(lang_code='b') # 'b' = British English
text = "Hello from Forge PC. This voice is running on my own computer."
generator = pipeline(text, voice='bf_emma')
for i, (gs, ps, audio) in enumerate(generator):
sf.write(f'{i}.wav', audio, 24000)
Run it with python speak.py. Kokoro downloads the model the first time, then saves the speech as WAV files at 24,000Hz.
The lang_code must match the voice. Kokoro uses 'a' for American English and 'b' for British English. British voices listed on the project's voices page include bf_alice, bf_emma, bf_isabella and bf_lily (female) and bm_daniel, bm_fable, bm_george and bm_lewis (male). Of these, bf_emma has the highest overall grade (B-) on that page.
Watch out for fake sites. The Kokoro Hugging Face page warns that websites using the Kokoro name are not owned by or affiliated with the project and are likely scams. Install it only from the official GitHub, Hugging Face or PyPI pages.
Option 2: Install and use Piper
Step 1: Install Piper
pip install piper-tts
Step 2: Download a voice
Run python -m piper.download_voices to list the available voices, then download one by name. Piper's documentation uses an American voice as its example:
python -m piper.download_voices en_US-lessac-medium
Piper's voice list includes English (Great Britain) voices with names starting en_GB, so pick one of those if you want a British accent. Voices download to the current folder unless you use --data-dir.
Step 3: Generate speech
python -m piper -m en_US-lessac-medium -f test.wav -- "This is a test."
This writes test.wav with the spoken sentence. Piper's documentation notes the command line is slow for repeated use because it loads the voice each time, and recommends its built-in web server instead. Other useful options include --input-file to read text from a file and --cuda for GPU acceleration, which requires the onnxruntime-gpu package.
The Open Home Foundation, which looks after Piper, says it's looking for maintainers for the project, so check the GitHub page for current status.
Tips
Read long documents in chunks. Both tools handle sentences and paragraphs well; split very long text into sections.
Check licences before commercial use. Kokoro's weights are Apache-licensed, while Piper is GPL-3.0 and each voice has its own licence.
Combine with a local AI model. You can pair local text-to-speech with a chatbot running in Ollama or LM Studio for a fully offline assistant. Our best free local AI apps guide lists more tools.
Our verdict
Kokoro and Piper make local text-to-speech free and private. Kokoro is the one to try first for natural-sounding speech, with British English voices from an Apache-licensed 82-million-parameter model, while Piper is lighter and faster for repeated use and smart home projects. Both install with pip on Windows 11 and run without a cloud account. Install them only from the official project pages; Forge PC hasn't tested either tool.