Whisper-Live-STT

OpenAI's whisper unfortunetly does not support live transcription. This repo is meant to fix that using a sliding window method.

Feel free to check the detailed article below on the explanations behind the sliding window method.

How to make Whisper STT live transcription. Part 1 Part 2 Part 3

Future Updates

• Create an installable package
• Create examples to create super class that can leverage custom models or custom audio streams.
• Fix the sentence builder algorithm, it deletes text when single words are cycled through.
• Add an iteractive gui

Installation

pip install -U openai-whisper
pip install pyaudio
pip install librosa
pip install numpy

Whisper requires ffmpeg, therefore the additional installation is required.

# on Ubuntu or Debian
sudo apt update && sudo apt install ffmpeg

# on Arch Linux
sudo pacman -S ffmpeg

# on MacOS using Homebrew (https://brew.sh/)
brew install ffmpeg

# on Windows using Chocolatey (https://chocolatey.org/)
choco install ffmpeg

# on Windows using Scoop (https://scoop.sh/)
scoop install ffmpeg

Alternatively, you can copy the ffmpeg binary file to the working directory of your project.

Usage/Examples

import os
import time

from live_stt_2 import LiveSTT

print("Initializing...")

stt = LiveSTT()
#stt.calculate_recommended_settings(5)

print("Running")

def clear_screen():
    # For Windows
    if os.name == 'nt':
        _ = os.system('cls')
    # For macOS and Linux
    else:
        _ = os.system('clear')

stt.start()

while True:
    time.sleep(2)
    clear_screen()

    print("Predicted Text:")
    print(stt.confirmed_text)

FAQ

FileNotFoundError: [WinError 2] The system cannot find the file specified

This error occurs when ffmpeg is not found in the environmental path or found in working directory. Check installation guide.