birdnet.acoustic.inference.core.encoding package

Submodules

birdnet.acoustic.inference.core.encoding.encoding_benchmarking module

class birdnet.acoustic.inference.core.encoding.encoding_benchmarking.FullBenchmarkEmbMeta(_start_timepoint, _end_timepoint, _time_wall_time_s, _file_durations, file_formats, mem_result_total_memory_usage_MiB, mem_shm_size_file_indices_MiB, mem_shm_size_segment_indices_MiB, mem_shm_size_audio_samples_MiB, mem_shm_size_batch_sizes_MiB, mem_shm_size_flags_MiB, file_segments_total, model_segment_duration_seconds, _time_rampup_first_line_s, sw_start_method, model_type, model_backend, model_version, model_is_custom, model_path, model_species, model_sig_fmin, model_sig_fmax, model_sample_rate, model_precision, file_segments_maximum, file_batches_processed, param_producers, param_workers, param_overlap_seconds, param_batch_size, param_prefetch_ratio, param_bandpass_fmin, param_bandpass_fmax, param_half_precision, param_devices, param_inference_library, worker_busy_average, worker_wait_time_average_milliseconds, speed_worker_xrt, speed_worker_xrt_max, _worker_avg_wall_time_s, mem_shm_ringsize, mem_memory_usage_maximum_MiB, mem_memory_usage_average_MiB, cpu_usage_maximum_pct, cpu_usage_average_pct, mem_shm_slots_average_free, mem_shm_slots_average_busy, mem_shm_slots_average_buffered, model_emb_dim)

Bases: FullBenchmarkMetaBase

model_emb_dim: int

class birdnet.acoustic.inference.core.encoding.encoding_benchmarking.MinimalBenchmarkEmbMeta(_start_timepoint, _end_timepoint, _time_wall_time_s, _file_durations, file_formats, mem_result_total_memory_usage_MiB, mem_shm_size_file_indices_MiB, mem_shm_size_segment_indices_MiB, mem_shm_size_audio_samples_MiB, mem_shm_size_batch_sizes_MiB, mem_shm_size_flags_MiB, file_segments_total, model_segment_duration_seconds)

Bases: MinimalBenchmarkMetaBase

birdnet.acoustic.inference.core.encoding.encoding_result module

class birdnet.acoustic.inference.core.encoding.encoding_result.AcousticDataEncodingResult(tensor, input_durations, segment_duration_s, overlap_duration_s, speed, model_path, model_fmin, model_fmax, model_sr, model_precision, model_version)

Bases: AcousticEncodingResultBase

class birdnet.acoustic.inference.core.encoding.encoding_result.AcousticEncodingResultBase(inputs, input_durations, model_path, model_fmin, model_fmax, model_sr, model_precision, model_version, segment_duration_s, overlap_duration_s, speed, tensor)

Bases: AcousticResultBase

property emb_dim: int

Return the embedding dimensionality.

Returns:

Number of coefficients per embedding vector.

Return type:

int

property embeddings: ndarray

Return the raw embedding tensor produced by the encoder.

Returns:

Embeddings with shape (n_inputs, n_segments, emb_dim).

Return type:

np.ndarray

property embeddings_masked: ndarray

Return the mask that marks relevant segments across files.

Returns:

Boolean mask of the same shape as embeddings.

Return type:

np.ndarray

property max_n_segments: int

Return the maximum segment count reserved per input.

Returns:

Number of overlapping windows available per file.

Return type:

int

property memory_size_MiB: float

Return the total result memory usage including embeddings buffers.

Returns:

Memory size in mebibytes.

Return type:

float

to_arrow_table()

Produce a PyArrow table that serializes each embedding with timing metadata.

Return type:

Table

Returns:

Table containing dictionary-encoded inputs and embeddings lists.

Return type:

pa.Table

to_csv(path, *, encoding='utf-8', buffer_size_kb=1024, silent=False)

Dump the structured embeddings to a CSV file for downstream analysis.

Return type:

None

Parameters:

• path – File path where the CSV will be written (must end with .csv).

• encoding – Text encoding for the output file.

• buffer_size_kb – Buffer size used when writing the file.

• silent – Suppress progress messages when True.

to_structured_array()

Convert the embeddings and timing metadata into a structured array.

Return type:

ndarray

Returns:

Array with fields for input path, start/end times, and embedding.

Return type:

np.ndarray

unprocessable_inputs()

Return the indices of inputs that could not be processed.

Return type:

ndarray

Returns:

Boolean mask or indices for skipped inputs.

Return type:

np.ndarray

class birdnet.acoustic.inference.core.encoding.encoding_result.AcousticFileEncodingResult(tensor, files, file_durations, segment_duration_s, overlap_duration_s, speed, model_path, model_fmin, model_fmax, model_sr, model_precision, model_version)

Bases: AcousticEncodingResultBase

birdnet.acoustic.inference.core.encoding.encoding_tensor module

class birdnet.acoustic.inference.core.encoding.encoding_tensor.AcousticEncodingTensor(session_id, n_inputs, emb_dim, half_precision, input_indices_dtype, segment_indices_dtype, max_segment_index)

Bases: AcousticTensorBase

copy_file_slice(file_idx, n_segments)

Return an independent copy of a single file’s embedding rows.

The returned arrays have shape (1, n_segments, emb_dim) so they can back a single-file result. Copying (rather than viewing) makes the data safe to hand to another thread while this tensor keeps being written/resized.

Must be called from the same thread that writes the tensor (the consumer); n_segments must not exceed the segments already written for the file.

Return type:

tuple[ndarray, ndarray]

property current_n_segments: int

property memory_usage_mb: float

set_unprocessable_inputs(unprocessable_inputs)

Return type:

None

write_block(file_indices, segment_indices, emb)

Return type:

None

class birdnet.acoustic.inference.core.encoding.encoding_tensor.PrebuiltEncodingTensor(emb, emb_masked)

Bases: AcousticTensorBase

Minimal tensor holder wrapping already-materialised per-file embeddings.

Used to build a single-file AcousticFileEncodingResult from the slice copied out of the shared result tensor, without re-running any inference.

property memory_usage_mb: float

write_block(*args, **kwargs)

Return type:

None

birdnet.acoustic.inference.core.encoding.encoding_worker module

class birdnet.acoustic.inference.core.encoding.encoding_worker.EncodingWorker(session_id, backend_loader, batch_size, n_slots, rf_file_indices, rf_segment_indices, rf_audio_samples, rf_batch_sizes, rf_flags, segment_duration_samples, out_q, wkr_ring_access_lock, sem_free, sem_fill, sem_active_workers, half_precision, wkr_stats_queue, logging_queue, logging_level, device, cancel_event, prd_all_done_event, start_signal, finish_signal, end_event, start_method)

Bases: WorkerBase

Module contents