jarvis-ui 0.7

Jarvis UI to perform voice commands via API calls

Deployments

Jarvis UI

Connects to Jarvis running in the backend to process request and response via API calls.

Kick off

Install

python -m pip install jarvis-ui

Initiate

import jarvis_ui


if __name__ == '__main__':
    jarvis_ui.start()

Mandatory Env Vars

• REQUEST_URL: URL to which the API call has to be made. Can be localhost or a tunneled URL behind a reverse proxy/CDN.
• TOKEN: Authentication token.

Optional Env Vars

• DEBUG: Defaults to False - Enable debug level logging
  
  
• REQUEST_TIMEOUT: Defaults to 5 - Timeout for API calls
• SPEECH_TIMEOUT: Defaults to 0 for macOS, 5 for Windows - Timeout for speech synthesis
  
  
• NATIVE_AUDIO: Defaults to False - If set to True, the response is generated in the server's default voice
• WAKE_WORDS: Defaults to jarvis (Defaults to alexa in macOS older than 10.14) - Wake words to initiate Jarvis
• SENSITIVITY: Defaults to 0.5 - Sensitivity of wake word detection
  
  
• MICROPHONE_INDEX: Defaults to None - Use peripherals.py to get the index values
• VOICE_NAME: Defaults to the author's favorite per the OS. Name of the voice supported by the OperatingSystem
• VOICE_RATE: Defaults to the value from pyttsx3 module - Speed/rate at which the text should be spoken
  
  
• VOICE_TIMEOUT: Defaults to 3 - Timeout for listener once wake word is detected - Awaits for a speech to begin until this limit
• VOICE_PHRASE_LIMIT: Defaults to None - Timeout for phrase once listener is activated - Listener will be deactivated after this limit
  
  
• RESTART_TIMER: Defaults to 24 hours.
  • I have a proxy tunneling my localhost from Jarvis server.
  • I have a CDN setup that's allowing me to access the tunnel with a custom domain.
  • Since I don't want my CDN to take the hit, and the tunnel origin changes often, I'm swapping the URL upon startup.
  • This requires restarting the UI every now and then.

Custom settings for speech recognition

These are customized according to the author's voice pitch. Please use test_listener.py to figure out the suitable values in a trial and error method.

These settings are added (optionally), to avoid the hard coded VOICE_PHRASE_LIMIT
Cons in using hard coded VOICE_PHRASE_LIMIT:

• Disables the listener after the set limit even when the speaker is actively talking.
• Listener will be active until the set limit even after the speaker has stopped talking.

Sample settings (formatted as JSON object)

• RECOGNIZER_SETTINGS: '{"energy_threshold": 1100, "dynamic_energy_threshold": false, "pause_threshold": 1, "phrase_threshold": 0.1}'

Description

• energy_threshold: Minimum audio energy to consider for recording. Greater the value, louder the speech should be.
• dynamic_energy_threshold: Change considerable audio energy threshold dynamically.
• pause_threshold: Seconds of non-speaking audio before a phrase is considered complete.
• phrase_threshold: Minimum seconds of speaking audio before it can be considered a phrase - values below this are ignored. This helps to filter out clicks and pops.
• non_speaking_duration: Seconds of non-speaking audio to keep on both sides of the recording.

Refer Jarvis' README for more information on setting up the backend server.

Coding Standards

Docstring format: Google
Styling conventions: PEP 8
Clean code with pre-commit hooks: flake8 and isort

Release Notes

Requirement

python -m pip install changelog-generator

Usage

changelog reverse -f release_notes.rst -t 'Release Notes'

Linting

PreCommit will ensure linting, and the doc creation are run on every commit.

Requirement

pip install --no-cache sphinx==5.1.1 pre-commit recommonmark

Usage

pre-commit run --all-files

Pypi Package

https://pypi.org/project/jarvis-ui/

Runbook

https://thevickypedia.github.io/Jarvis_UI/

License & copyright

© Vignesh Sivanandha Rao

Licensed under the MIT License