doty

Doty, take this down...

Doty is (primarily) a Mumble <> IRC bridge/transcription bot. It will transcribe any speech in a Mumble channel and output it to IRC, while copying or speaking any IRC messages into Mumble. Currently only English is supported, though the various libraries doty uses could support other languages so that might be improved in the future.

Doty provides a couple of options for Speech-to-Text but the quality of the transcription regardless of engine is merely passable. Automated transcription does not seem to be geared toward casual conversation so it can make a lot of mistakes, but generally you can reasonably follow a conversation in Mumble just from the transcription.

Pre-requisites

Doty no longer requires any 3rd party web service accounts for speech, but still requires some preparation:

General Commands

• WolframAlpha: A WolframAlpha API key is required for the WolframAlpha query intent
• YouTube API: A YouTube API key is required to search videos on YouTube

Speech to Text Engines

• vosk: A model will need to be downloaded and pointed to in the configuration. The transcription is relatively good, but does not support hint phrases so recognition on names can be very spotty.
• coqui-stt: A model and scorer will need to be downloaded and pointed to in the configuration. Not recommended, the transcription is very poor and slow.
• ibm-watson: An IBM Watson Cloud account with an IAM API key authorized to use the Speech-to-Text service. The transcription is reasonably good but even casual use will probably be rather expensive.

Text to Speech Engines

• coqui-tts: The configured model will be automatically downloaded when doty starts up. The quality of the speech is (usually) quite good but can have some occasional hiccups, and is very slow.
• larynx-tts: A model and vocoder will need to be downloaded, extracted and pointed to in the configuration. Trades a little quality for speed compared to coqui-tts.
• google-tts: A Google Cloud account with a JSON blob of credentials to use the Text-to-Speech service. The quality of the speech is good, and casual use will probably fall under the free-tier.

Installation

Build/run requirements:

• python-devel
• libsamplerate
• rust
• cargo
• gcc-c++
• ffmpeg
• youtube-dl
• rnnoise

To install (recommend using a virtualenv):

pip install -Ur requirements.pre-install.txt
pip install -Ur requirements.txt
snips-nlu download en

Usage

Edit a copy of the example config file and set it up to your tastes:

cp -a config.yml.example my-config.yml
$EDITOR my-config.yml

Then you can just run doty against that config file:

./doty.py -c my-config.yml

Doty has several commandline options, though most of them are only useful for debugging:

usage: doty.py [-h] [-c CONFIG] [-v] [-q] [-M] [-I] [-T] [-S] [-A]

optional arguments:
  -h, --help            show this help message and exit
  -c CONFIG, --config CONFIG
                        Path to config file
  -v, --verbose         Log more messages
  -q, --quiet           Log fewer messages
  -M, --no-mumble       Do not start the mumble process
  -I, --no-irc          Do not start the irc process
  -T, --no-transcriber  Do not start the transcriber process
  -S, --no-speaker      Do not start the speaker process
  -A, --no-autorestart  Don't automatically restart crashed processes

Commands

Doty supports commands through voice or text (in mumble or IRC), there are examples of the available voice commands in command-dataset.yml, though they will need to be prefixed by a configured activation word (ex: "robot, help").

"robot, what can you do?"

You can tell me to: make an audio clip or replay audio, change to a different channel, or just ask a question. You can also tell me to play radio stations and change the volume

!help

Available commands: !help, !say, !play-ytdl, !play