Functions — github.com/ASLP-lab/SenSE

hub / github.com/ASLP-lab/SenSE / functions

Functions360 in github.com/ASLP-lab/SenSE

⨍ Functions 360◇ Types & classes 78

↓ 41 callersMethodload

(cls, model_path)

src/sense/model/encoder/wavlm/wavlm_encoder.py:14

↓ 29 callersMethodupdate

(self, cfg: dict)

src/sense/model/encoder/wavlm/wavlm.py:218

↓ 18 callersFunctionexists

(v)

src/sense/model/utils.py:30

↓ 12 callersMethod__init__

(self, dim)

src/sense/model/modules.py:357

↓ 8 callersFunctionpadded_mel_batch

(ref_mels)

src/sense/eval/utils_eval.py:78

↓ 7 callersFunctiondefault

(v, d)

src/sense/model/utils.py:34

↓ 6 callersMethod__init__

Construct an MultiHeadedAttention object.

src/sense/model/modules2.py:89

↓ 6 callersMethodsample

( self, cond: float["b n d"] | float["b nw"], # noqa: F722 cond_noisy: float["b n d"]

src/sense/model/cfm.py:88

↓ 5 callersMethod__init__

(self, channels, se_bottleneck_dim=128)

src/sense/eval/ecapa_tdnn.py:75

↓ 5 callersMethodgenerate

(self, mels, mels_len, wavs=None, wavs_len=None, ref_mels=None, ref_mels_len=None, ref_tokens=None, ref_tokens

src/sense/model/llama_llm.py:149

↓ 5 callersFunctionget_metainfo_w_prompt

(metalst)

src/sense/eval/utils_eval.py:20

↓ 5 callersFunctionload_vocoder

(vocoder_name="vocos", is_local=False, local_path="", device=device, hf_cache_dir=None)

src/sense/infer/utils_infer.py:99

↓ 5 callersFunctionnormal_

(data)

src/sense/model/modules2.py:180

↓ 5 callersFunctionrepetition_found

(text, length=2, tolerance=10)

src/sense/model/utils.py:210

↓ 5 callersFunctionsetup_encoder

(model_name, model_path=None, freeze_encoder=False, **kwargs)

src/sense/model/encoder/encoder_wrapper.py:5

↓ 4 callersMethod__init__

( self, in_channels: int, out_channels: int, kernel_size: int,

src/sense/model/encoder/conformer/convolution.py:43

↓ 4 callersFunctionload_dataset

dataset_type - "CustomDataset" if you want to use tokenizer name and default data path to load for train_dataset - "Custom

src/sense/model/dataset.py:366

↓ 4 callersFunctionload_fm_model

( model_cls, model_cfg, mel_spec_cfg, ckpt_path, # mel_spec_type=mel_spec_type, ode_me

src/sense/infer/utils_infer.py:250

↓ 4 callersFunctionmake_pad_mask

Make mask tensor containing indices of padded part. See description of make_non_pad_mask. Args: lengths (torch.Tensor): Batch of len

src/sense/model/utils.py:78

↓ 4 callersFunctionpadded_wav_batch

(wavs)

src/sense/eval/utils_eval.py:88

↓ 4 callersFunctionquant_noise

Wraps modules and applies quantization noise to the weights for subsequent quantization with Iterative Product Quantization as described

src/sense/model/modules2.py:201

↓ 3 callersMethod__init__

( self, cfg: WavLMConfig, )

src/sense/model/encoder/wavlm/wavlm.py:223

↓ 3 callersMethod__init__

(self, shape: tuple, contiguous: bool = False)

src/sense/model/encoder/conformer/modules.py:54

↓ 3 callersFunctionadjust_noise_length

Adjust noise length to match target audio length. If noise is longer, truncate it. If noise is shorter, repeat it.

src/sense/train/noisy_datasets/prepare_libritts_noisy.py:47

↓ 3 callersFunctionconvert_char_to_pinyin

(text_list, polyphone=True)

src/sense/model/utils.py:163

↓ 3 callersFunctionget_pos_embed_indices

(start, length, max_pos, scale=1.0)

src/sense/model/modules.py:341

↓ 3 callersFunctionget_tokenizer

tokenizer - "pinyin" do g2p for only chinese characters, need .txt vocab_file - "char" for char-wise tokenizer, need .txt vocab

src/sense/model/utils.py:127

↓ 3 callersFunctionlens_to_mask

(t: int["b"], length: int | None = None)

src/sense/model/utils.py:41

↓ 3 callersFunctionload_llm_model

( model_cls, model_cfg, ckpt_path, device=device )

src/sense/infer/utils_infer.py:240

↓ 3 callersFunctionprecompute_freqs_cis

(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0)

src/sense/model/modules.py:327

↓ 3 callersMethodsave_checkpoint

(self, update, last=False)

src/sense/model/trainer_llm.py:143

↓ 3 callersMethodsave_checkpoint

(self, update, last=False)

src/sense/model/trainer_cfm.py:149

↓ 3 callersMethodtrain

(self, train_dataset: Dataset, num_workers=16, resumable_with_seed: int = None)

src/sense/model/trainer_llm.py:258

↓ 2 callersMethod__init__

(self, ds_rate, encoder_dim, llm_dim)

src/sense/model/projector.py:29

↓ 2 callersMethod__init__

( self, *, dim, depth=8, heads=8, dim_head=64, dropout

src/sense/model/backbones/mmdit.py:85

↓ 2 callersMethod__init__

( self, *, dim, depth=8, heads=8, dim_head=64, dropout

src/sense/model/backbones/unett.py:107

↓ 2 callersMethod__init__

( self, *, dim, depth=8, heads=8, dim_head=64, dropout

src/sense/model/backbones/dit.py:105

↓ 2 callersMethod_add_bos_eos

(self, bos, eos, inputs_embeds, attention_mask, target=None, ignore_target=True)

src/sense/model/llama_llm.py:218

↓ 2 callersFunctionadd_noise_to_audio

Add noise to clean audio with specified SNR. Args: clean_audio: Clean speech signal noise_audio: Noise signal (same leng

src/sense/train/noisy_datasets/prepare_libritts_noisy.py:66

↓ 2 callersMethodapply_mask

(self, x, padding_mask)

src/sense/model/encoder/wavlm/wavlm.py:273

↓ 2 callersMethodbackward

(ctx, grad)

src/sense/model/modules2.py:67

↓ 2 callersFunctioncompute_mask_indices

Computes random mask spans for a given shape Args: shape: the the shape for which to compute masks. should be of size 2

src/sense/model/encoder/wavlm/wavlm.py:37

↓ 2 callersMethodextend_pe

(self, x: Tensor)

src/sense/model/encoder/conformer/embedding.py:35

↓ 2 callersMethodextract_variable_length_features

x : torch.Tensor, shape = (batch_size, n_mels, n_ctx) the mel spectrogram of the audio

src/sense/model/encoder/whisper/whisper_encoder.py:13

↓ 2 callersMethodforward_padding_mask

( self, features: torch.Tensor, padding_mask: torch.Tensor, )

src/sense/model/encoder/wavlm/wavlm.py:313

↓ 2 callersMethodget_embedding_from_mel

return: wav_embedding: (b, l, v) wav_mask: (b, l), positions with valid wav values are true

src/sense/model/llama_llm.py:184

↓ 2 callersFunctionget_inference_prompt

( metainfo, target_sample_rate=24000, n_fft=1024, win_length=1024, n_mel_channels=100,

src/sense/eval/utils_eval.py:100

↓ 2 callersFunctionget_metainfo_wo_prompt

(metalst)

src/sense/eval/utils_eval.py:30

↓ 2 callersMethodget_speech_token_label_embedding

(self, speech_token_labels, speech_tokens_length)

src/sense/model/llama_llm.py:210

↓ 2 callersFunctionis_chinese

(c)

src/sense/model/utils.py:173

↓ 2 callersFunctionload_checkpoint

(model, ckpt_path, device: str, dtype=None, use_ema=True)

src/sense/infer/utils_infer.py:193

↓ 2 callersFunctionload_noise_list

Load noise file list from .scp file.

src/sense/train/noisy_datasets/prepare_libritts_noisy.py:22

↓ 2 callersFunctionmask_from_frac_lengths

(seq_len: int["b"], frac_lengths: float["b"])

src/sense/model/utils.py:57

↓ 2 callersFunctionpreprocess_audio

(audio_path_or_array, sample_rate=16000, n_mels=128)

src/sense/model/encoder/whisper/whisper_encoder.py:42

↓ 2 callersMethodprocess_feats

Args: audio (torch.Tensor): Audio waveform tensor (1, T).

src/sense/eval/SpeechBERTScore/speechbertscore.py:75

↓ 2 callersMethodset_epoch

Sets the epoch for this sampler.

src/sense/model/dataset.py:343

↓ 2 callersFunctionsetup_encoder_projector

(encoder_projector, **kwargs)

src/sense/model/projector_wrapper.py:5

↓ 1 callersFunctionECAPA_TDNN_SMALL

( feat_dim, emb_dim=256, feat_type="wavlm_large", sr=16000, feature_selection="hidden_stat

src/sense/eval/ecapa_tdnn.py:313

↓ 1 callersMethod__init__

( self, encoder_dim: int = 512, num_attention_heads: int = 8,

src/sense/model/encoder/conformer/encoder2.py:55

↓ 1 callersMethod__init__

( self, d_model: int = 512, num_heads: int = 16, dropout_p: fl

src/sense/model/encoder/conformer/attention.py:46

↓ 1 callersMethod__init__

( self, encoder_dim: int = 512, num_attention_heads: int = 8,

src/sense/model/encoder/conformer/encoder.py:55

↓ 1 callersMethod__init__

(self, dim: int)

src/sense/model/encoder/conformer/activation.py:36

↓ 1 callersFunction_amp_to_db

(x, min_level_db)

src/sense/model/modules.py:94

↓ 1 callersMethod_append_prev_key_padding_mask

( key_padding_mask: Optional[Tensor], prev_key_padding_mask: Optional[Tensor],

src/sense/model/modules2.py:764

↓ 1 callersMethod_get_input_buffer

( self, incremental_state: Optional[Dict[str, Dict[str, Optional[Tensor]]]] )

src/sense/model/modules2.py:807

↓ 1 callersFunction_normalize

(S, max_abs_value, min_db)

src/sense/model/modules.py:99

↓ 1 callersMethod_relative_positions_bucket

(self, relative_positions, bidirectional=True)

src/sense/model/modules2.py:415

↓ 1 callersMethod_relative_shift

(self, pos_score: Tensor)

src/sense/model/encoder/conformer/attention.py:105

↓ 1 callersMethod_set_input_buffer

( self, incremental_state: Dict[str, Dict[str, Optional[Tensor]]], buffer:

src/sense/model/modules2.py:817

↓ 1 callersFunctionadd_noise_to_audio

Add noise to clean audio with specified SNR. Args: clean_audio: Clean speech signal noise_audio: Noise signal (same leng

src/sense/train/noisy_datasets/process_one_file.py:66

↓ 1 callersFunctionadjust_noise_length

Adjust noise length to match target audio length. If noise is longer, truncate it. If noise is shorter, repeat it.

src/sense/train/noisy_datasets/process_one_file.py:47

↓ 1 callersMethodapply_sparse_mask

(self, attn_weights, tgt_len: int, src_len: int, bsz: int)

src/sense/model/modules2.py:824

↓ 1 callersFunctionarrange

(s, e, length, keep_length)

src/sense/model/encoder/wavlm/wavlm.py:112

↓ 1 callersMethodaudio_melspec

(self, audio, n_mels=120, frame_size=320, hop_length=160, sr=16000, to_db=True)

src/sense/eval/eval_dnsmos.py:23

↓ 1 callersFunctionbatch_convert_texts

Convert a list of texts to pinyin in batches.

src/sense/train/datasets/prepare_csv_wavs.py:82

↓ 1 callersFunctionbert_score

Args: v_generated (torch.Tensor): Generated feature tensor (T, D). v_reference (torch.Tensor): Reference feature tensor (T, D).

src/sense/eval/SpeechBERTScore/speechbertscore.py:16

↓ 1 callersFunctioncalculate_intrusive_score

(gt_dir, testset_dir, csv_path=None, json_path=None)

src/sense/eval/intrusive_se_metrics.py:80

↓ 1 callersMethodckpt_wrapper

(self, module)

src/sense/model/backbones/dit.py:175

↓ 1 callersMethodclear_cache

(self)

src/sense/model/backbones/dit.py:183

↓ 1 callersFunctioncli

()

src/sense/train/datasets/prepare_csv_wavs.py:245

↓ 1 callersMethodcompute_bias

(self, query_length, key_length)

src/sense/model/modules2.py:442

↓ 1 callersFunctioncopy_directory_structure

Copy directory structure and files efficiently, skipping files that already exist

src/sense/train/noisy_datasets/prepare_libritts_noisy.py:168

↓ 1 callersFunctioncopy_tree_optimized

(src, dst)

src/sense/train/noisy_datasets/prepare_libritts_noisy.py:215

↓ 1 callersMethodcount_parameters

Count parameters of encoder

src/sense/model/encoder/conformer/encoder2.py:172

↓ 1 callersMethodcount_parameters

Count parameters of encoder

src/sense/model/encoder/conformer/encoder.py:171

↓ 1 callersMethodestoi_metric

(self, ref, inf, fs=16000)

src/sense/eval/intrusive_se_metrics.py:29

↓ 1 callersMethodextract_features

( self, source: torch.Tensor, padding_mask: Optional[torch.Tensor] = None, mas

src/sense/model/encoder/wavlm/wavlm.py:325

↓ 1 callersMethodextract_features

(self, x, padding_mask=None, streaming_mask=None, tgt_layer=None)

src/sense/model/encoder/wavlm/wavlm.py:622

↓ 1 callersMethodextract_features

(self, source, padding_mask)

src/sense/model/encoder/wavlm/wavlm_encoder.py:36

↓ 1 callersFunctionget_activation_fn

Returns the activation function corresponding to `activation`

src/sense/model/modules2.py:142

↓ 1 callersFunctionget_args

()

src/sense/eval/eval_metrics.py:20

↓ 1 callersFunctionget_audio_duration

Get the duration of an audio file in seconds using ffmpeg's ffprobe. Falls back to torchaudio.load() if ffprobe fails.

src/sense/train/datasets/prepare_csv_wavs.py:157

↓ 1 callersFunctionget_audio_files

(custom_dir, prompt_dir=None)

src/sense/eval/utils_eval.py:40

↓ 1 callersMethodget_feat

(self, x)

src/sense/eval/ecapa_tdnn.py:271

↓ 1 callersMethodget_feat_num

(self)

src/sense/eval/ecapa_tdnn.py:260

↓ 1 callersMethodget_frame_len

(self, index)

src/sense/model/dataset.py:127

↓ 1 callersMethodget_polyfit_val

(self, sig, bak, ovr)

src/sense/eval/eval_dnsmos_oval.py:29

↓ 1 callersMethodget_polyfit_val

(self, sig, bak, ovr)

src/sense/eval/eval_dnsmos.py:29

↓ 1 callersFunctionget_speech_enhancement_test

(metalst, gen_wav_dir, gpus, eval_ground_truth=False)

src/sense/eval/utils_eval.py:348

↓ 1 callersFunctiongraceful_exit

Context manager for graceful shutdown on signals

src/sense/train/datasets/prepare_csv_wavs.py:46

next →1–100 of 360, ranked by callers