Workflow

• Capture audio input.
• Convert audio to text.
• Send text to local Hugging Face Models.
• Retrieve and process the intent and parameters.
• Query the SQLite database.
• Convert the result to audio output.
• Play the response to the user.

Dependencies

• Linux
• alsa-utils - arecord (version 1.2.13) is used for voice recording from the microphone. Other alternative is PortAudio (cross-compatible)
• espeak-ng- espeak (version 1.52) is used for text to speech. Other alternatives are festival.
• Assembly.AI account - assembly.ai is used for speech to text. Other alternatives are Google Speech to Text.
• Python - A virtual environment should be created. The directory is venv. Install requirements from requirements.txt on the virtual environment. The first time calling scripts/main.py could take a while since some resources need to be download - subsequent calls are faster. The models in scripts/custom_models.py should be trained prior. Based on the results, the classification_model & ner_model paths can be updated in scripts/process_intent_and_parameters
• Go - All dependencies should be downloaded. Data should be seed into the database on the initial build/run.

Data source

• Covid Cases
• Covid Management Protocol. ChatGPT is used to summarize the document.

Query Types

Case-Based Queries

• Cases by Date: "How many active cases are in Karnataka today?"
• Max Cases: "What’s the highest number of recoveries in Delhi this week?"
• Average Cases: "Show me the average deaths in Maharashtra over the past 30 days."
• Total Cases: "How many cases were reported in Goa last month?"

Trend-Based Queries

• Location-Based: "Which state has the highest number of active cases?"
• Date-Based: "When did Tamil Nadu cross 1,000 active cases for the first time?"

Freeform Questions

Ask anything related to COVID-19 data

Resources

• ASCII Generator