STTS | Wikispeech

Wikimedia's official information about the project: www.mediawiki.org/wiki/Wikispeech

1. Installation

These instructions are currently being updated, and is a work in progress (Q1, 2026)

1.1 Software

Below is a list of components to install for Wikispeech. Go to each git repo (linked below), and follow installation instructions in the README file.

Symbolset Build Status Symbolset/mapping server Tested commit ID: 3d7fb68
Pronlex Build Status Lexicon server Tested commit ID: 3130d31
Matcha TTS Matcha server Adapter for Matcha TTS Tested commit ID: 3e234e6
Piper TTS Piper server Adapter for Piper TTS Tested commit ID: 3e234e6
Wikispeech-server Build Status The main Wikispeech-server Tested commit ID: b9cd3f4

1.2 Download models for Matcha and Piper

1.2.1 Deep Phonemizer models

From STTS' Google Drive folder, download the following file to $HOME/.local/share/deep_phonemizer:

1.2.2 Matcha models

    mkdir -p $HOME/.local/share/matcha_tts
    cd $HOME/.local/share/matcha_tts
    curl -L https://github.com/shivammehta25/Matcha-TTS-checkpoints/releases/download/v1.0/generator_v1 -o hifigan_T2_v1
    curl -L https://github.com/shivammehta25/Matcha-TTS-checkpoints/releases/download/v1.0/g_02500000 -o hifigan_univ_v1
    curl -L https://github.com/shivammehta25/Matcha-TTS-checkpoints/releases/download/v1.0/matcha_ljspeech.ckpt -o matcha_ljspeech.ckpt
    curl -L https://github.com/shivammehta25/Matcha-TTS-checkpoints/releases/download/v1.0/matcha_vctk.ckpt -o matcha_vctk.ckpt
From STTS' Google Drive folder, download the following files to $HOME/.local/share/matcha_tts:

1.2.3 Piper models

    mkdir -p $HOME/.local/share/piper_tts
    cd $HOME/.local/share/piper_tts
    wget https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/ar/ar_JO/kareem/medium/ar_JO-kareem-medium.onnx
    wget https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/ar/ar_JO/kareem/medium/ar_JO-kareem-medium.onnx.json
    wget https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/en/en_US/bryce/medium/en_US-bryce-medium.onnx
    wget https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/en/en_US/bryce/medium/en_US-bryce-medium.onnx.json
Alignment patching for Arabic and US English: 
    cd <WIKISPEECH-TTS-WRAPPERS>/piper_server
    source .venv/bin/activate
    uv pip install onnx
    python3 -m piper.patch_voice_with_alignment $HOME/.local/share/piper_tts/en_US-bryce-medium.onnx
    python3 -m piper.patch_voice_with_alignment $HOME/.local/share/piper_tts/ar_JO-kareem-medium.onnx
From STTS' Google Drive folder, download the following files to $HOME/.local/share/piper_tts:

2. Running Wikispeech

Convenience scripts for running the full standard setup with all servers:
If you want to start each server separately, please see each repo's README file for details. You can also have a look at the start_wikispeech.sh script for clues. Please note that the sub-servers must be up and running before the main Wikispeech server can be started.

2. API documentation

When you have installed and started the Wikispeech server, you can find the API documentation in the URLs below. (Examples for servers located on localhost, with default ports.)

Main API/Wikispeech server: http://localhost:10000

Sub-servers:

3. Release notes

4. System requirements

RAM: 4-8G

The following may also be worth to keep in mind:

Wikispeech has been tested for the following specific OSs:

5. Adding new components/voices/languages to Wikispeech

Instructions will be updated in 2026

At the moment, there is no comprehensive documentation available on how to build a new voice or a new language for Wikispeech.

If you want to contribute to Wikispeech, please consult with the Wikimedia team who will coordinate community based improvements. You can read more about how to contribute here: www.mediawiki.org/wiki/Wikispeech.

You are also welcome to contact us at stts.se/contact if you want.

5.1 Symbol set

Documentation of the symbolset format

5.2 Lexicon data

Documentation of the Wikispeech lexicon file format

A lexicon file in this format can be imported using the lexserver API URL /admin/lex_import_page, or from the command line using importLex.

5.4 Configuring Wikispeech for a new voice/language

The wikispeech server is currently configured with a python source file "voice_config.py", containing two lists, textprocessor_configs and voice_configs. To add a new language, add entries for textprocessor and voice, similar to the examples below. Please note that the examples here use the MaryTTS adapter. 
Textprocessor:
    {"name":"TEXTPROCESSOR_NAME", "lang":"MYLANGAUGE",
     "components":[
         {
             "module":"adapters.marytts_adapter",
             "call":"marytts_preproc",
             "mapper": {
                 "from":"LEXICON_SYMBOLSET",
                 "to":"MARYTTS_SYMBOLSET"
             },
         },
         {
             "module":"adapters.lexicon_client",
             "call":"lexLookup",
             "lexicon":"MY_LEXICON_NAME"
         }
     ]
    }
	    

Voice:
    {
        "lang":"MYLANGUAGE",
        "name":"MARYTTS_VOICE_NAME",
        "engine":"marytts",
        "adapter":"adapters.marytts_adapter",
        "mapper": {
            "from":"LEXICON_SYMBOL_SET",
            "to":"MARYTTS_SYMBOL_SET"
            }
    }

6. More information

An UML diagram over the Wikispeech server: github.com/stts-se/wikispeech-server/wiki/uml/basic.png

Github repositories

Main repos

Self-financed repos

A list of all STTS' public repositories on github: stts-se.github.io/repos


This work was supported by the Swedish Post and Telecom Authority (PTS) through the grant "Wikispeech – en användargenererad talsyntes på Wikipedia" (2016–2017), and by Swedish Inheritance Fund ("Allmänna Arvsfonden") through the grant "Wikispeech Talsyntes och taldatainsamlare" (2024–2026).

© STTS Södermalms talteknologiservice 2018-2026 | stts.se