YiliVoice - Voice-to-Text Tool

A personal development tool that uses OpenAI's Whisper model for responsive voice recognition and transcription.

Features

• Real-time voice-to-text transcription powered by a background Whisper worker
• Minimal overlay indicator to show capture state at a glance
• GPU acceleration support (CUDA) when available
• Smart duplicate and filler filtering with a customizable block list
• Automatic low-volume gating to avoid transcribing background noise
• Drops high no-speech probability outputs to reduce "thank you" hallucinations
• Auto-shutdown after extended inactivity with quick restart

Requirements

• Python 3.7+
• CUDA-compatible GPU (optional, for faster processing)

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/yiliVoice.git
   cd yiliVoice

2. Install required packages:

   pip install numpy speech_recognition openai-whisper torch pyautogui keyboard tkinter

Usage

Run the script with default settings:

python yiliVoice.py

Command Line Arguments

• --model: Choose Whisper model size (tiny/base/small/medium/large) [default: medium]
• --non_english: Enable non-English language detection
• --energy_threshold: Microphone detection sensitivity [default: 1000]
• --record_timeout: Audio buffer length in seconds [default: 1.5]
• --phrase_timeout: Silence duration that ends a phrase in seconds [default: 1.0]
• --volume_threshold: Minimum peak volume (0-1 range) to keep audio [default: 0.008]
• --trailing_silence: Additional silence captured after speech stops [default: 0.8]
• --threshold_adjustment: Multiplier that makes repetition filtering stricter [default: 1.0]
• --no_speech_threshold: Suppress output when Whisper predicts high no-speech probability [default: 0.6]

Controls

• Ctrl+F8: Toggle recording on/off
• Indicator colors: Green = recording, Pink = ready, Red = auto-shutdown

Status Indicator

A small overlay circle appears at the top-center of your screen:

• Green indicates active recording
• Pink shows the system is ready
• Red appears when the system enters auto-shutdown mode after roughly 10 minutes of inactivity

Features in Detail

Automatic Filtering

The system automatically filters out:

• Duplicate phrases
• Partial repetitions
• Common filler phrases (customizable in the filter list)
• Short phrases below the minimum word count
• High no-speech probability segments that usually produce "thank you" hallucinations

Performance

• Dedicated worker threads for audio capture, transcription, and UI updates
• Bytearray-backed audio buffer to minimize copying and memory churn
• GPU acceleration when available (CUDA)
• Ambient noise calibration during start-up