The nvidia riva SDK requires a generator for realtime STT - so we have to wrap the

Internal: populate the simulator verdict / run before on_simulation_end.

Build an ITU-T G.711 A-law to linear 16-bit PCM lookup table. Note: A-law sign convention is inverted relative to mu-law -- when the sign bit

Build an ITU-T G.711 mu-law to linear 16-bit PCM lookup table.

Construct, once, a virtual-clock validator for every model with a captured real clock.

Build a JSON-serializable tool schema with full parameter type info.

Build WebSocket URL with parameters.

Calculate audio duration from buffer.

Calculate session duration based on credential expiry and AWS 8-min limit.

Release gate for a verdict (which verdict it is, is decided elsewhere). - post-speech silence is required for every verdict - end-of-

Capture file-descriptor-level output for the duration of a concurrent gather. The per-task :class:`_CaptureController` only swaps `sys.stdout`/`s

Convert chat context to langgraph input

Warn about deprecated kwargs and migrate values where possible.

Check the energy state of an audio frame.

Check for error in RTZR response and raise if found.

Declare a programmatic user-driven turn. Pins ``user_state`` to ``"speaking"`` and holds ``wait_for_idle`` open until release. On rel