Convert to string (lowercase)

Error

Get immutable access to the underlying buffer

Log a message asynchronously (non-blocking)

Push new audio chunk to the buffer

Get current backend

Stream implementation

Set current backend

Get human-readable name

Get a new database connection

Clear all buffers in the pool

Flush any remaining audio and return final speech segments

Add an item to be processed in a batch

Get mutable access to the underlying buffer

Spawn the sidecar process

Load notification settings from disk

Wait for the next processed result

Get the current model name

* Initialize database connection

Consume the wrapper and return the buffer (will not be returned to pool)

Run the VAD-driven audio processing pipeline

Get sample rate

Stop the stream

Create transcript segments from transcription results. Each tuple is (text, start_ms, end_ms) from VAD timestamps.

Show a notification if all conditions are met

Find audio file in meeting folder Tries common names first, then scans for any file with an audio extension

Get a specific model by name

Get speech chunks with progress callback and cancellation support The callback receives (progress_percent, segments_found) and returns false to cancel

Save notification settings to disk

Strips the first `#` heading line; returns "" if no `#` is found.

Convert decoded audio to Whisper-compatible 16kHz mono f32 format. Performs mono conversion, normalization, and resampling. Large files (>5 min at 48

Create guard and set flag atomically

Get adaptive buffer timeout range for this device type Returns (min_timeout, max_timeout) in milliseconds These values are based on: - Cap's buffer

Resample large audio files in fixed-size chunks through the sinc resampler. Processes `input` in 60-second chunks using the high-quality sinc resampl

Generate synthetic speech-like audio with alternating speech/silence

Load and parse a template by identifier This function implements a fallback strategy: 1. Check user's custom templates directory 2. Check bundled res

Check if the engine has a model loaded

Process incoming audio samples and return any complete speech segments Handles resampling from input sample rate to 16kHz for VAD processing

Discover available Parakeet models.

Generates a summary using the specified LLM provider # Arguments `client` - Reqwest HTTP client (reused for performance) `provider` - The LLM provide

Get a buffer from the pool, or create a new one if pool is empty