Installation on Arch Linux

[DarkTyger]

Overview

These instructions describe how to perform speech-to-text transcription
using Arch Linux.

Hardware

Your setup will differ.

Components

Hardware components:

• Mixer = Mackie d.4 PRO
• Mic = PANASONIC Density Mic
• Motherboard = ROG STRIX B650E-I GAMING WIFI

Configuration

Hardware setup (numbers in brackets are from figures in manual):

• Mic -> Mixer (Mic 1 [2], 48V off)
• Mixer (AUX SEND [8]) -> Motherboard (Mic In)
• Mic -> On
• Mixer (Mic 1) -> [26] Wet (100%)
• Mixer (Mic 1) -> [40] AUX SEND (75%)
• Mixer (Mic 1) -> [28] On

This configuration allows audio to be sent from the microphone to the
computer's mic input without any audio going out mains into the speakers.
Adjusting the Wet / Dry level controls how much audio is split to mains.

Install

This section provides instructions to run the speech-to-text software.

Refresh audio

To avoid audio issues, consider completely reinstalling audio software:

rm -rf $HOME/.config/alsa
rm -rf $HOME/.config/pavucontrol

sudo rm -rf /etc/alsa
sudo rm -rf /etc/pulse
sudo rm -rf /var/lib/alsa

sudo pacman -Sy \
  pulseaudio pulseaudio-alsa \
  alsa-lib alsa-tools alsa-plugins alsa-utils alsa-firmware 

The audio environment is configured.

Speech-to-text

Change to the project's root directory and perform the following steps:

• Install cuda, cudnn, and patchelf
• Install alsa-lib, alsa-utils, and jack2
• Install Python 3.10.9 in local environment
• Install ffmpeg-python
• Install RealtimeSTT

For example:

sudo pacman -S cuda cudnn patchelf
sudo pacman -S alsa-lib alsa-utils jack2
sudo pacman -S pyenv
pyenv install 3.10.9

Change to your project's development directory (e.g., $HOME/dev/c++/project)
and run the following:

pyenv local 3.10.9

pip install --upgrade pip
pip install ffmpeg-python
pip install RealtimeSTT

patchelf --clear-execstack $HOME/.pyenv/versions/3.10.9/lib/python3.10/site-packages/ctranslate2.libs/libctranslate2-bc15bf3f.so.4.5.0

The requisite software is installed.

Configure

Append the following lines to $HOME/.bashrc:

export PYENV_ROOT="$HOME/.pyenv"
eval "$(pyenv init - bash)"

Open a new terminal and close all other terminals to ensure the environment
variables are applied.

Demo

Create and run the demo program by saving the following as stt.py:

$ cat tts.py
from RealtimeSTT import AudioToTextRecorder

def process_text(text):
    print(text)

if __name__ == '__main__':
    print("Wait until it says 'speak now'")
    recorder = AudioToTextRecorder()

    while True:
        recorder.text(process_text)

From the shell, run:

python stt.py

Speech is translated to text.