Initializes a PipelineRequest instance. Args: action: The type of action requested (e.g., "prepare", "abort").

Initializes a RunningGeneration state object. Args: id: A unique identifier for this generation attempt.

Initializes the SpeechPipelineManager. Sets up configuration, instantiates dependencies (AudioProcessor, LLM, etc.), loads s

Background thread worker to check for abort conditions based on partial text.

Worker thread target that handles LLM inference for a generation. Waits for `generator_ready_event`. Once signaled, it iterates thro

Internal callback relay triggered when the transcriber starts recording.

Worker thread target that processes requests from the `requests_queue`. Continuously monitors the queue. When a request arrives, it

Internal callback relay for silence detection status.

Background worker thread that processes text from the queue for turn detection. Continuously retrieves text items from `self.text_qu

Worker thread target that handles TTS synthesis for the 'final' part of the answer. Continuously checks the `running_generation`. It

Worker thread target that handles TTS synthesis for the 'quick answer'. Waits for `llm_answer_ready_event`. Once signaled, it checks

Clears the cache of potentially yielded sentences. This effectively stops any further actions that might be triggered based

Finds and attempts to cancel requests older than the specified timeout. Iterates through active requests and calls `cancel_generatio

Serves the favicon.ico file. Returns: A FileResponse containing the favicon.

Public method to signal the end of user input or interaction. Queues a 'finish' action onto the `requests_queue`. Note: Curr

Returns the final remaining segment of upsampled audio after all chunks are processed. After the last call to `get_base64_chunk`, th

Formats the log record's creation time into MM:SS.cs format. Uses `time.localtime` to convert the record's creation timestamp and

Yields remaining text chunks for final TTS synthesis.

Serves the main index.html page. Reads the content of static/index.html and returns it as an HTML response. Returns: An HTMLRes

Gets the response for a requested path, adding no-cache headers. Args: path: The path to the static file requested.

Manages the application's lifespan, initializing and shutting down resources. Initializes global components like SpeechPipelineManager, Upsa

Callback executed when the RealtimeTTS audio stream stops processing. Logs the event and sets the `finished_event` to signal complet

Callback invoked just before the final STT result for a user turn is confirmed. Sets flags indicating user finished, allows TTS if p

Callback invoked when the final transcription result for a user turn is available. Logs the final transcription and stores it.

Callback triggered for real-time transcription updates.

Callback invoked when a partial transcription result is available. Updates internal state, sends the partial result to the client,

Callback invoked if the STT detects a potential need to abort based on user speech.

Callback invoked when a potential *final* transcription is detected (hot state). Logs the potential final transcription. Ar

Callback invoked when a potentially complete sentence is detected by the STT. Triggers the preparation of a speech generation based

Callback invoked when the audio input processor starts recording user speech. If client-side TTS is playing, it triggers an interrup

Callback invoked when the silence detection state changes. Updates the internal silence_active flag. Args: sile

Handles partial transcription results from the transcriber.

Manually triggers the final transcription process using the last known real-time text. This bypasses the recorder's natural

Resets the pipeline state completely. Aborts any currently running generation (waiting for completion) and clears the conver

Placeholder for safely aborting syntheses (currently does nothing).

Shuts down the recorder instance, cleans up resources, and prevents further processing. Sets the `shutdown_performed` flag.

Initiates a graceful shutdown of the pipeline manager and worker threads. 1. Sets the `shutdown_event`. 2. Attempts a final

Callback triggered when recorder starts a new recording segment.

Callback triggered when recorder detects start of silence (end of speech).

Callback triggered when recorder stops a recording segment, just before final transcription might be generated.

Callback triggered when recorder detects end of silence (start of speech).

Sets up the final transcription callback mechanism with the recorder. This method defines the `on_final` callback that will be invok

Handles the main WebSocket connection for real-time voice chat. Accepts a connection, sets up connection-specific state via `TranscriptionCa