Voice to text Telegram bot

Extensible Telegram bot for converting voice messages to text.

It uses Vosk API for speech recognition (https://alphacephei.com/vosk/)

Start the bot using Docker compose

First create a bot using BotFather (https://core.telegram.org/bots#3-how-do-i-create-a-bot)

In the project root directory create .env file and put your bot token there:

TG_BOT_TOKEN=...

Then run:

docker compose up

Working example

Working example can be found here: https://t.me/xiovbot
Though it is hosted with a free plan and cannot process more then one voice message at a time due to low RAM.

How to add a new language

Bot currently supports two languages: English and Russian.
You can extend bot functionality by adding a new language.
For example, suppose you want to add Spanish, then in the conf.py file add a new entry to LANGUAGES dict:

LANGUAGES = {
    ...,
    'es': {
        'full': _('spanish'),
        'emoji': '🇪🇸'
    },
}

Then specify name of the model you want to use in MODELS:

MODELS = {
    ...,
    'es': 'vosk-model-small-es-0.22',
}

List of all available models you can find here: https://alphacephei.com/vosk/models or https://alphacephei.com/vosk/models/model-list.json
After that you need to create .po file in locales/LANG/LC_MESSAGES/ where LANG is the language code, es in our example:

mkdir -p locales/es/LC_MESSAGES/
msginit --no-translator -i locales/bot.pot -o locales/es/LC_MESSAGES/bot.po

msginit command requires gettext package installed. You can install it with apt-get update && apt-get install gettext (Ubuntu)

Then edit newly created bot.po manually to provide bot interface translation, you may keep it untouched and all interface will remain in English.

And compile bot.po to .mo:

msgfmt locales/es/LC_MESSAGES/bot.po -o locales/es/LC_MESSAGES/bot.mo