API Reference

Model Class

aic.Model(model_type=AICModelType.QUAIL_L, license_key=None, *, sample_rate, channels=1, frames=None, allow_variable_frames=False)

RAII + context-manager convenience around the C interface.

Parameters:

• model_type (AICModelType, default: QUAIL_L ) –

  The neural model variant to load; defaults to :pydata:AICModelType.QUAIL_L.

• license_key (str | bytes, default: None ) –

  Optional signed license string. Empty string means trial mode.

Create a model wrapper.

Parameters:

• model_type (AICModelType, default: QUAIL_L ) –

  The neural model variant to load; defaults to :pydata:AICModelType.QUAIL_L.

• license_key (str | bytes, default: None ) –

  Signed license string. Required. Obtain a key at https://developers.ai-coustics.io.

• sample_rate (int) –

  Input/output sample rate in Hz. Required.

• channels (int, default: 1 ) –

  Channel count. Optional, defaults to 1.

• frames (int | None, default: None ) –

  Optional block length in frames for streaming. If omitted, the model's :py:meth:optimal_num_frames will be used.

close()

Explicitly free native resources (idempotent).

create_vad()

Create a Voice Activity Detector bound to this model.

get_parameter(param)

Get the current value of a parameter.

Parameters:

• param (AICParameter | AICEnhancementParameter) –

  Parameter enum value. See :py:class:aic._bindings.AICEnhancementParameter.

Returns:

• float –

  The current value of the parameter.

library_version() staticmethod

Return the version string of the underlying AIC SDK library.

Returns:

• str –

  Semantic version string.

optimal_num_frames()

Return the suggested buffer length for streaming.

Returns:

• int –

  Recommended block size in frames.

optimal_sample_rate()

Return the suggested I/O sample rate for the loaded model.

Returns:

• int –

  Sample rate in Hz.

process(pcm, *, channels=None)

Enhance pcm in-place using planar processing (convenience pass-through).

Parameters:

• pcm (ndarray) –

  Planar 2-D array of shape (channels, frames) Data must be float32 in the linear -1…+1 range. Any non-conforming array is copied to a compliant scratch buffer.

• channels (int | None, default: None ) –

  Override channel count auto-detected from pcm. Rarely needed.

Returns:

• ndarray –

  The same array instance (modified in-place) or a contiguous copy if a dtype/stride conversion had been necessary.

process_async(pcm, *, channels=None) async

Async variant of :py:meth:process executed on the model's worker thread.

process_interleaved(pcm, channels)

Enhance pcm in-place using interleaved processing (convenience pass-through).

Parameters:

• pcm (ndarray) –

  Interleaved 1-D array of shape (frames,) containing interleaved audio data Data must be float32 in the linear -1…+1 range. Any non-conforming array is copied to a compliant scratch buffer.

• channels (int) –

  Number of channels in the interleaved data.

Returns:

• ndarray –

  The same array instance (modified in-place) or a contiguous copy if a dtype/stride conversion had been necessary.

process_interleaved_async(pcm, channels) async

Async variant of :py:meth:process_interleaved executed on the model's worker thread.

process_interleaved_submit(pcm, channels)

Submit :py:meth:process_interleaved to the worker thread, returning a Future.

process_sequential(pcm, channels)

Enhance pcm in-place using sequential channel data processing.

Processes audio where all samples for each channel are stored sequentially (channel 0 samples, then channel 1 samples, etc.) rather than interleaved.

Parameters:

• pcm (ndarray) –

  Sequential 1-D array of shape (frames * channels,) containing sequential audio data where all samples for channel 0 come first, followed by all samples for channel 1, etc. Data must be float32 in the linear -1…+1 range. Any non-conforming array is copied to a compliant scratch buffer.

• channels (int) –

  Number of channels in the sequential data.

Returns:

• ndarray –

  The same array instance (modified in-place) or a contiguous copy if a dtype/stride conversion had been necessary.

process_sequential_async(pcm, channels) async

Async variant of :py:meth:process_sequential executed on the model's worker thread.

process_sequential_submit(pcm, channels)

Submit :py:meth:process_sequential to the worker thread, returning a Future.

process_submit(pcm, *, channels=None)

Submit :py:meth:process to the model's worker thread, returning a Future.

processing_latency()

Return the current output delay (in samples).

Returns:

• int –

  End-to-end delay in samples at the configured sample rate.

reset()

Flush the model's internal state (between recordings, etc.).

set_parameter(param, value)

Update an algorithm parameter.

Parameters:

• param (AICParameter | AICEnhancementParameter) –

  Parameter enum value. See :py:class:aic._bindings.AICEnhancementParameter.

• value (float) –

  New value for the parameter (float).

Raises:

• RuntimeError –

  If the parameter is out of range or the SDK call fails.

Voice Activity Detector

aic.VoiceActivityDetector(model)

Voice Activity Detector bound to a :pyclass:Model.