Upload 14 files

CKPT.yaml

@@ -0,0 +1,4 @@
+# yamllint disable
+end-of-epoch: true
+si-snr: -7.369511792677621
+unixtime: 1621680451.6093628

README.md

@@ -0,0 +1,117 @@
+---
+language: "en"
+thumbnail:
+tags:
+- audio-to-audio 
+- audio-source-separation
+- Source Separation
+- Speech Separation
+- WHAM!
+- SepFormer
+- Transformer 
+- pytorch
+- speechbrain
+license: "apache-2.0"
+metrics:
+- SI-SNRi
+- SDRi
+
+---
+
+<iframe src="https://ghbtns.com/github-btn.html?user=speechbrain&repo=speechbrain&type=star&count=true&size=large&v=2" frameborder="0" scrolling="0" width="170" height="30" title="GitHub"></iframe>
+<br/><br/>
+
+# SepFormer trained on WHAMR! (16k sampling frequency)
+This repository provides all the necessary tools to perform audio source separation with a [SepFormer](https://arxiv.org/abs/2010.13154v2) model, implemented with SpeechBrain, and pretrained on [WHAMR!](http://wham.whisper.ai/) dataset with 16k sampling frequency, which is basically a version of WSJ0-Mix dataset with environmental noise and reverberation in 16k. For a better experience we encourage you to learn more about [SpeechBrain](https://speechbrain.github.io). The given model performance is 13.5 dB SI-SNRi on the test set of WHAMR! dataset.
+
+
+| Release | Test-Set SI-SNRi | Test-Set SDRi |
+|:-------------:|:--------------:|:--------------:|
+| 30-03-21 | 13.5 dB | 13.0 dB |
+
+
+## Install SpeechBrain
+
+First of all, please install SpeechBrain with the following command:
+
+```
+pip install speechbrain
+```
+
+Please notice that we encourage you to read our tutorials and learn more about [SpeechBrain](https://speechbrain.github.io).
+
+### Perform source separation on your own audio file
+
+```python
+from speechbrain.inference.separation import SepformerSeparation as separator
+import torchaudio
+
+model = separator.from_hparams(source="speechbrain/sepformer-whamr16k", savedir='pretrained_models/sepformer-whamr16k')
+
+# for custom file, change path
+est_sources = model.separate_file(path='speechbrain/sepformer-whamr16k/test_mixture16k.wav') 
+
+torchaudio.save("source1hat.wav", est_sources[:, :, 0].detach().cpu(), 16000)
+torchaudio.save("source2hat.wav", est_sources[:, :, 1].detach().cpu(), 16000)
+```
+
+The system expects input recordings sampled at 16kHz (single channel).
+If your signal has a different sample rate, resample it (e.g, using torchaudio or sox) before using the interface.
+
+### Inference on GPU
+To perform inference on the GPU, add  `run_opts={"device":"cuda"}`  when calling the `from_hparams` method.
+
+### Training
+The model was trained with SpeechBrain (fc2eabb7).
+To train it from scratch follows these steps:
+1. Clone SpeechBrain:
+```bash
+git clone https://github.com/speechbrain/speechbrain/
+```
+2. Install it:
+```
+cd speechbrain
+pip install -r requirements.txt
+pip install -e .
+```
+
+3. Run Training:
+```
+cd  recipes/WHAMandWHAMR/separation/
+python train.py hparams/sepformer-whamr.yaml --data_folder=your_data_folder --sample_rate=16000
+```
+
+You can find our training results (models, logs, etc) [here](https://drive.google.com/drive/folders/1QiQhp1vi5t4UfNpNETA48_OmPiXnUy8O?usp=sharing).
+
+### Limitations
+The SpeechBrain team does not provide any warranty on the performance achieved by this model when used on other datasets.
+
+#### Referencing SpeechBrain
+
+```bibtex
+@misc{speechbrain,
+  title={{SpeechBrain}: A General-Purpose Speech Toolkit},
+  author={Mirco Ravanelli and Titouan Parcollet and Peter Plantinga and Aku Rouhe and Samuele Cornell and Loren Lugosch and Cem Subakan and Nauman Dawalatabad and Abdelwahab Heba and Jianyuan Zhong and Ju-Chieh Chou and Sung-Lin Yeh and Szu-Wei Fu and Chien-Feng Liao and Elena Rastorgueva and François Grondin and William Aris and Hwidong Na and Yan Gao and Renato De Mori and Yoshua Bengio},
+  year={2021},
+  eprint={2106.04624},
+  archivePrefix={arXiv},
+  primaryClass={eess.AS},
+  note={arXiv:2106.04624}
+}
+```
+
+
+#### Referencing SepFormer
+```bibtex
+@inproceedings{subakan2021attention,
+      title={Attention is All You Need in Speech Separation}, 
+      author={Cem Subakan and Mirco Ravanelli and Samuele Cornell and Mirko Bronzi and Jianyuan Zhong},
+      year={2021},
+      booktitle={ICASSP 2021}
+}
+```
+
+# **About SpeechBrain**
+- Website: https://speechbrain.github.io/
+- Code: https://github.com/speechbrain/speechbrain/
+- HuggingFace: https://huggingface.co/speechbrain/

brain.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9e24193f36931b7f57932532efbdcf64971f42732383ba6808825f77db258f6
+size 28

config.json

@@ -0,0 +1,3 @@
+{
+  "speechbrain_interface": "SepformerSeparation"
+}

counter.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dbb1ded63bc70732626c5dfe6c7f50ced3d560e970f30b15335ac290358748f6
+size 3

dataloader-TRAIN.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39e5b4830d4d9c14db7368a95b65d5463ea3d09520373723430c03a5a453b5df
+size 5

decoder.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d959cb46de5b15f008bf5476cf7c19bc680b310c7e10883e3eddfdbde533cb8
+size 17272

encoder.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31d23d395a408b887b8f6ac01e477f00bcba27f95785426359fb52d52f1dc6ed
+size 17272

gitattributes

@@ -0,0 +1,17 @@
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tar.gz filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text

hyperparams.yaml

@@ -0,0 +1,65 @@
+# ################################
+# Model: Inference for source separation with SepFormer
+# https://arxiv.org/abs/2010.13154
+# Generated from speechbrain/recipes/WSJ0Mix/separation/train/hparams/sepformer-whamr-16khz.yaml
+# Dataset : Whamr-16kHz
+# ###############################
+
+
+# Parameters
+sample_rate: 16000
+num_spks: 2
+
+# Specifying the network
+Encoder: !new:speechbrain.lobes.models.dual_path.Encoder
+  kernel_size: 16
+  out_channels: 256
+
+SBtfintra: !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
+  num_layers: 8
+  d_model: 256
+  nhead: 8
+  d_ffn: 1024
+  dropout: 0
+  use_positional_encoding: true
+  norm_before: true
+
+SBtfinter: !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
+  num_layers: 8
+  d_model: 256
+  nhead: 8
+  d_ffn: 1024
+  dropout: 0
+  use_positional_encoding: true
+  norm_before: true
+
+MaskNet: !new:speechbrain.lobes.models.dual_path.Dual_Path_Model
+  num_spks: !ref <num_spks>
+  in_channels: 256
+  out_channels: 256
+  num_layers: 2
+  K: 250
+  intra_model: !ref <SBtfintra>
+  inter_model: !ref <SBtfinter>
+  norm: ln
+  linear_layer_after_inter_intra: false
+  skip_around_intra: true
+
+Decoder: !new:speechbrain.lobes.models.dual_path.Decoder
+  in_channels: 256
+  out_channels: 1
+  kernel_size: 16
+  stride: 8
+  bias: false
+
+modules:
+  encoder: !ref <Encoder>
+  decoder: !ref <Decoder>
+  masknet: !ref <MaskNet>
+
+pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
+  loadables:
+    masknet: !ref <MaskNet>
+    encoder: !ref <Encoder>
+    decoder: !ref <Decoder>
+

hyperparams_train.yaml

@@ -0,0 +1,184 @@
+# Generated 2021-05-22 from:
+# /home/mila/s/subakany/speechbrain_new/recipes/WSJ0Mix/separation/yamls/sepformer-whamr-16k.yaml
+# yamllint disable
+# ################################
+# Model: SepFormer for source separation
+# https://arxiv.org/abs/2010.13154
+#
+# Dataset : WSJ0-2mix and WSJ0-3mix
+# ################################
+# Basic parameters
+# Seed needs to be set at top of yaml, before objects with parameters are made
+#
+seed: 1234
+__set_seed: !apply:torch.manual_seed [1234]
+
+# Data params
+
+# the data folder for the wham dataset
+# data_folder needs to follow the format:  /yourpath/whamr.
+# make sure to use the name whamr at your top folder for the dataset!
+data_folder: /network/tmp1/subakany/whamr_16k
+
+# the path for wsj0/si_tr_s/ folder -- only needed if dynamic mixing is used
+# e.g. /yourpath/wsj0-processed/si_tr_s/
+# you need to convert the original wsj0 to 8k
+# you can do this conversion with the script ../meta/preprocess_dynamic_mixing.py
+wsj0_tr: /yourpath/wsj0-processed/si_tr_s/
+
+experiment_name: sepformer-whamr-randomreverb-16k
+output_folder: results/sepformer-whamr-randomreverb-16k/1234
+train_log: results/sepformer-whamr-randomreverb-16k/1234/train_log.txt
+save_folder: results/sepformer-whamr-randomreverb-16k/1234/save
+
+# the file names should start with whamr instead of whamorg
+train_data: results/sepformer-whamr-randomreverb-16k/1234/save/whamr_tr.csv
+valid_data: results/sepformer-whamr-randomreverb-16k/1234/save/whamr_cv.csv
+test_data: results/sepformer-whamr-randomreverb-16k/1234/save/whamr_tt.csv
+skip_prep: false
+
+# Experiment params
+auto_mix_prec: false # Set it to True for mixed precision
+test_only: true
+num_spks: 2 # set to 3 for wsj0-3mix
+progressbar: true
+save_audio: false # Save estimated sources on disk
+sample_rate: 16000
+
+# Training parameters
+N_epochs: 200
+batch_size: 1
+lr: 0.00015
+clip_grad_norm: 5
+loss_upper_lim: 999999  # this is the upper limit for an acceptable loss
+# if True, the training sequences are cut to a specified length
+limit_training_signal_len: true
+# this is the length of sequences if we choose to limit
+# the signal length of training sequences
+training_signal_len: 64000
+
+# Set it to True to dynamically create mixtures at training time
+dynamic_mixing: false
+
+# Parameters for data augmentation
+
+# rir_path variable points to the directory of the room impulse responses
+# e.g. /miniscratch/subakany/rir_wavs
+# If the path does not exist, it is created automatically.
+rir_path: /network/tmp1/subakany/rir_wavs_16k
+
+use_wavedrop: false
+use_speedperturb: true
+use_speedperturb_sameforeachsource: false
+use_rand_shift: false
+min_shift: -8000
+max_shift: 8000
+
+speedperturb: !new:speechbrain.lobes.augment.TimeDomainSpecAugment
+  perturb_prob: 1.0
+  drop_freq_prob: 0.0
+  drop_chunk_prob: 0.0
+  sample_rate: 16000
+  speeds: [95, 100, 105]
+
+wavedrop: !new:speechbrain.lobes.augment.TimeDomainSpecAugment
+  perturb_prob: 0.0
+  drop_freq_prob: 1.0
+  drop_chunk_prob: 1.0
+  sample_rate: 16000
+
+# loss thresholding -- this thresholds the training loss
+threshold_byloss: true
+threshold: -30
+
+# Encoder parameters
+N_encoder_out: 256
+out_channels: 256
+kernel_size: 16
+kernel_stride: 8
+
+# Dataloader options
+dataloader_opts:
+  batch_size: 1
+  num_workers: 3
+
+# Specifying the network
+Encoder: &id003 !new:speechbrain.lobes.models.dual_path.Encoder
+  kernel_size: 16
+  out_channels: 256
+
+
+SBtfintra: &id001 !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
+  num_layers: 8
+  d_model: 256
+  nhead: 8
+  d_ffn: 1024
+  dropout: 0
+  use_positional_encoding: true
+  norm_before: true
+
+SBtfinter: &id002 !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
+  num_layers: 8
+  d_model: 256
+  nhead: 8
+  d_ffn: 1024
+  dropout: 0
+  use_positional_encoding: true
+  norm_before: true
+
+MaskNet: &id005 !new:speechbrain.lobes.models.dual_path.Dual_Path_Model
+
+  num_spks: 2
+  in_channels: 256
+  out_channels: 256
+  num_layers: 2
+  K: 250
+  intra_model: *id001
+  inter_model: *id002
+  norm: ln
+  linear_layer_after_inter_intra: false
+  skip_around_intra: true
+
+Decoder: &id004 !new:speechbrain.lobes.models.dual_path.Decoder
+  in_channels: 256
+  out_channels: 1
+  kernel_size: 16
+  stride: 8
+  bias: false
+
+optimizer: !name:torch.optim.Adam
+  lr: 0.00015
+  weight_decay: 0
+
+loss: !name:speechbrain.nnet.losses.get_si_snr_with_pitwrapper
+
+lr_scheduler: &id007 !new:speechbrain.nnet.schedulers.ReduceLROnPlateau
+
+  factor: 0.5
+  patience: 2
+  dont_halve_until_epoch: 85
+
+epoch_counter: &id006 !new:speechbrain.utils.epoch_loop.EpochCounter
+  limit: 200
+
+modules:
+  encoder: *id003
+  decoder: *id004
+  masknet: *id005
+checkpointer: !new:speechbrain.utils.checkpoints.Checkpointer
+  checkpoints_dir: results/sepformer-whamr-randomreverb-16k/1234/save
+  recoverables:
+    encoder: *id003
+    decoder: *id004
+    masknet: *id005
+    counter: *id006
+    lr_scheduler: *id007
+train_logger: !new:speechbrain.utils.train_logger.FileTrainLogger
+  save_file: results/sepformer-whamr-randomreverb-16k/1234/train_log.txt
+
+
+pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
+   loadables:
+      masknet: !ref <MaskNet>
+      encoder: !ref <Encoder>
+      decoder: !ref <Decoder>

lr_scheduler.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5108af5b18f633bd908c3142c8e87f80f95d3159ba00cb2c52317cfd2d8e1b5
+size 1647

masknet.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5fb8c690668e5d1bbbc9a8256974577093a09cd08e845e9f30024fdd33472ce
+size 113112646

optimizer.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89a9b119d5ce268c345c2ba978e8ff10a13beabc8d6ee1962b256721de9564d6
+size 205694713