Persephone

README.md

@@ -26,13 +26,24 @@ The full API is described in the documentation page [https://hyperion-ml.readthe
 ### Prerequisites
 
     We use anaconda or miniconda, though you should be able to make it work in other python distributions
-    To start, you should create a new enviroment and install PyTorch>=1.9, (older versions are not supported any longer) e.g.:
+    To start, you should create a new enviroment and install PyTorch:
 ```
-conda create --name ${your_env} python=3.8
+conda create --name ${your_env} python=3.11
 conda activate ${your_env}
-conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=10.2 -c pytorch
+# We used PyTorch 2.0.1, other versions may work too
+conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.8 -c pytorch -c nvidia
+# If using k2 for ASR
+wget https://huggingface.co/csukuangfj/k2/resolve/main/ubuntu-cuda/k2-1.24.4.dev20240223+cuda11.8.torch2.0.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
+pip install k2-1.24.4.dev20240223+cuda11.8.torch2.0.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
 ```
-In next Hyperion versions, we will upgrade to Pytorch>=1.9 and drop compatibility with older PyTorch versions.
+
+For older systems with cuda 10.2 driver:
+```
+conda create --name ${your_env} python=3.10
+conda activate ${your_env}
+conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=10.2 -c pytorch
+```
+
 
 ### Installing Hyperion
 
@@ -41,27 +52,12 @@ In next Hyperion versions, we will upgrade to Pytorch>=1.9 and drop compatibilit
 git clone https://github.com/hyperion-ml/hyperion.git
 ```
 
-- You can choose to install hyperion in the environment
+- Then install hyperion in the environment
 ```bash
 cd hyperion
 pip install -e .
 ```
 
-- Or add the hyperion toolkit to the PYTHONPATH envirnoment variable
-  This option will allow you to share the same environment if you are working with several hyperion branches
-  at the same time, while installing it requires to have an enviroment per branch.
-  For this, you need to install the requirements
-```bash
-cd hyperion
-pip install -r requirements.txt
-```
-Then add these lines to your `~/.bashrc` or to each script that uses hyperion
-```bash
-HYP_ROOT= #substitute this by your hyperion location
-export PYTHONPATH=${HYP_ROOT}:$PYTHONPATH
-export PATH=${HYP_ROOT}/bin:$PATH
-```
-
 ## Recipes
 
 There are recipes for several tasks in the `./egs` directory.

apps.txt

@@ -1,69 +1,14 @@
-apply-mvn-select-frames.py
-compute-energy-vad.py
-compute-mfcc-feats.py
-copy-feats.py
-eval-cos-1vs1.py
-eval-linear-gbe-up.py
-eval-linear-gbe.py
-eval-linear-svmc.py
-eval-logistic-regression.py
-eval-plda-1vs1.py
-eval-plda-nvs1.py
-make-babble-noise-audio-files.py
-merge-h5-files.py
-pack-audio-files.py
-pack-wav-rirs.py
-plot-vector-hist.py
-plot-vector-tsne.py
-preprocess-audio-files.py
-rttm-to-bin-vad.py
-segments-to-bin-vad.py
-torch-adv-finetune-xvec-from-wav.py
-torch-adv-finetune-xvec.py
-torch-compute-mfcc-feats.py
-torch-eval-vae.py
-torch-eval-xvec-cosine-scoring-from-adv-test-wav-wavegan.py
-torch-eval-xvec-cosine-scoring-from-adv-test-wav.py
-torch-eval-xvec-cosine-scoring-from-art-test-wav.py
-torch-eval-xvec-cosine-scoring-from-test-wav.py
-torch-eval-xvec-cosine-scoring-from-transfer-adv-test-wav.py
-torch-eval-xvec-cosine-scoring-from-transfer-art-test-wav.py
-torch-eval-xvec-logits-from-wav.py
-torch-extract-xvectors-from-wav-with-rttm.py
-torch-extract-xvectors-from-wav.py
-torch-extract-xvectors-slidwin-from-wav.py
-torch-extract-xvectors-slidwin.py
-torch-extract-xvectors-vae-preproc.py
-torch-extract-xvectors.py
-torch-finetune-xvec-dfr-from-wav.py
-torch-finetune-xvec-dfr.py
-torch-finetune-xvec-from-wav.py
-torch-finetune-xvec.py
-torch-generate-adv-attacks-xvector-classif.py
-torch-generate-adv-attacks-xvector-verif.py
-torch-train-dvae.py
-torch-train-efficientnet-xvec-from-wav.py
-torch-train-efficientnet-xvec.py
-torch-train-resnet-xvec-from-wav.py
-torch-train-resnet-xvec.py
-torch-train-spinenet-xvec-from-wav.py
-torch-train-tdnn-xvec-from-wav.py
-torch-train-tdnn-xvec.py
-torch-train-transformer-xvec-v1-from-wav.py
-torch-train-transformer-xvec-v1.py
-torch-train-vae.py
-torch-train-vq-dvae.py
-torch-train-vq-vae.py
-torch-train-xvec-from-wav.py
-train-cw-up.py
-train-cw.py
-train-gaussianizer.py
-train-lda.py
-train-linear-gbe-up.py
-train-linear-gbe.py
-train-linear-svmc.py
-train-logistic-regression.py
-train-mvn.py
-train-nda.py
-train-pca.py
-train-plda.py
+compute_energy_vad.py
+extract_wav2vec2xvectors.py
+extract_xvectors_from_wav.py
+finetune_wav2vec2xvector.py
+finetune_xvector_dfr_from_feats.py
+finetune_xvector_dfr_from_wav.py
+finetune_xvector_from_feats.py
+finetune_xvector_from_wav.py
+make_babble_noise_audio_files.py
+pack_wav_rirs.py
+preprocess_audio_files.py
+train_wav2vec2xvector.py
+train_xvector_from_feats.py
+train_xvector_from_wav.py

egs/chime5_spkdet/v1/local/score_dcf.py

@@ -15,7 +15,7 @@
 from hyperion.hyp_defs import float_cpu, config_logger
 from hyperion.utils.trial_scores import TrialScores
 from hyperion.utils.trial_key import TrialKey
-from hyperion.metrics import fast_eval_dcf_eer as fast_eval
+from hyperion.np.metrics import fast_eval_dcf_eer as fast_eval
 
 
 def score_dcf(key_file, score_file, output_path):

egs/chime5_spkdet/v1/steps_be/eval-be-diar-v2.py

@@ -17,7 +17,7 @@
 from hyperion.utils import TrialScores
 from hyperion.helpers import MultiTestTrialDataReaderV2 as TDR
 from hyperion.helpers import PLDAFactory as F
-from hyperion.transforms import TransformList
+from hyperion.np.transforms import TransformList
 
 
 def combine_diar_scores(ndx, orig_seg, subseg_scores):

egs/chime5_spkdet/v1/steps_be/eval-be-v1.py

@@ -19,7 +19,7 @@
 from hyperion.utils.trial_scores import TrialScores
 from hyperion.helpers import TrialDataReader as TDR
 from hyperion.helpers import PLDAFactory as F
-from hyperion.transforms import TransformList
+from hyperion.np.transforms import TransformList
 
 
 def eval_plda(

egs/chime5_spkdet/v1/steps_be/eval-calibration-v1.py

@@ -18,8 +18,8 @@
 from hyperion.utils.trial_scores import TrialScores
 from hyperion.utils.trial_key import TrialKey
 from hyperion.utils.trial_ndx import TrialNdx
-from hyperion.metrics import compute_act_dcf, compute_min_dcf
-from hyperion.classifiers import BinaryLogisticRegression as LR
+from hyperion.np.metrics import compute_act_dcf, compute_min_dcf
+from hyperion.np.classifiers import BinaryLogisticRegression as LR
 
 
 def eval_calibration(in_score_file, ndx_file, model_file, out_score_file):

egs/chime5_spkdet/v1/steps_be/train-be-v1.py

@@ -17,7 +17,7 @@
 from hyperion.hyp_defs import float_cpu, config_logger
 from hyperion.helpers import VectorClassReader as VCR
 from hyperion.helpers import VectorReader as VR
-from hyperion.transforms import TransformList, LDA, LNorm
+from hyperion.np.transforms import TransformList, LDA, LNorm
 from hyperion.helpers import PLDAFactory as F
 
 

egs/chime5_spkdet/v1/steps_be/train-calibration-v1.py

@@ -17,8 +17,8 @@
 from hyperion.hyp_defs import float_cpu, config_logger
 from hyperion.utils.trial_scores import TrialScores
 from hyperion.utils.trial_key import TrialKey
-from hyperion.metrics import compute_act_dcf, compute_min_dcf
-from hyperion.classifiers import BinaryLogisticRegression as LR
+from hyperion.np.metrics import compute_act_dcf, compute_min_dcf
+from hyperion.np.classifiers import BinaryLogisticRegression as LR
 
 
 def train_calibration(score_file, key_file, model_file, prior, verbose):

egs/dihard2019/v1/steps_diar/eval-ahc-v1.py

@@ -27,13 +27,13 @@
 from hyperion.io import RandomAccessDataReaderFactory as DRF
 from hyperion.io import VADReaderFactory as VRF
 from hyperion.helpers import PLDAFactory as F
-from hyperion.transforms import TransformList, PCA, LNorm
-from hyperion.clustering import AHC
-from hyperion.pdfs import GMMTiedDiagCov as GMM
-from hyperion.diarization import DiarAHCPLDA as Diar
+from hyperion.np.transforms import TransformList, PCA, LNorm
+from hyperion.np.clustering import AHC
+from hyperion.np.pdfs import GMMTiedDiagCov as GMM
+from hyperion.np.diarization import DiarAHCPLDA as Diar
 
-# from hyperion.pdfs import GMMDiagCov as GMM2
-# from hyperion.pdfs import GMM as GMM3
+# from hyperion.np.pdfs import GMMDiagCov as GMM2
+# from hyperion.np.pdfs import GMM as GMM3
 
 
 def make_timestamps(n, win_start, win_length, win_shift, win_shrink):

egs/dihard2019/v1/steps_diar/train-plda-v1.py

@@ -22,7 +22,7 @@
 from hyperion.utils import Utt2Info
 
 # from hyperion.helpers import VectorClassReader as VCR
-from hyperion.transforms import TransformList, LDA, LNorm, PCA
+from hyperion.np.transforms import TransformList, LDA, LNorm, PCA
 from hyperion.helpers import PLDAFactory as F
 from hyperion.io import RandomAccessDataReaderFactory as DRF
 

egs/librispeech/v0/cmd.sh

@@ -0,0 +1,28 @@
+# you can change cmd.sh depending on what type of queue you are using.
+# If you have no queueing system and want to run on a local machine, you
+# can change all instances 'queue.pl' to run.pl (but be careful and run
+# commands one by one: most recipes will exhaust the memory on your
+# machine).  queue.pl works with GridEngine (qsub).  slurm.pl works
+# with slurm.  Different queues are configured differently, with different
+# queue names and different ways of specifying things like memory;
+# to account for these differences you can create and edit the file
+# conf/queue.conf to match your queue's configuration.  Search for
+# conf/queue.conf in http://kaldi-asr.org/doc/queue.html for more information,
+# or search for the string 'default_config' in utils/queue.pl or utils/slurm.pl.
+
+if [ "$(hostname -d)" == "cm.gemini" ];then
+    export train_cmd="queue.pl --config conf/coe_gpu_long.conf --mem 4G"
+    export cuda_cmd="queue.pl --config conf/coe_gpu_long.conf --mem 20G"
+    #export cuda_cmd="queue.pl --config conf/coe_gpu_v100.conf --mem 20G"
+    export cuda_cmd="queue.pl --config conf/coe_gpu_rtx.conf --mem 40G"
+    export cuda_eval_cmd="queue.pl --config conf/coe_gpu_short.conf --mem 4G"
+    #export cuda_eval_cmd="queue.pl --config conf/coe_gpu_rtx.conf --mem 10G"
+    #export cuda_eval_cmd="queue.pl --config conf/coe_gpu_long.conf --mem 4G"
+else
+    export train_cmd="queue.pl --config conf/clsp.conf --mem 4G " 
+    export cuda_cmd="queue.pl --config conf/clsp.conf --mem 20G"
+    export cuda_eval_cmd="$train_cmd"
+fi
+
+
+

egs/librispeech/v0/conf/clsp.conf

@@ -0,0 +1,11 @@
+
+# Default configuration
+command qsub -v PATH -cwd -S /bin/bash -j y -l arch=*64* -V
+option mem=* -l mem_free=$0,ram_free=$0
+option mem=0          # Do not add anything to qsub_opts
+option num_threads=* -pe smp $0
+option num_threads=1  # Do not add anything to qsub_opts
+option max_jobs_run=* -tc $0
+default gpu=0
+option gpu=0 -l 'hostname=b[1]*|c0[123456789]*|c1[1345679]*|c2[12357]*'
+option gpu=* -l 'hostname=c0[123456789]*|c1[1345679]*|c2[12357]*,gpu=$0'

egs/librispeech/v0/conf/infer.yaml

@@ -0,0 +1,4 @@
+beam_width: 5
+decoding_method: time_sync_beam_search
+#decoding_method: greedy
+#decoding_method: align_length_sync_beam_search

egs/librispeech/v0/conf/reverb_noise20dB_aug.yaml

@@ -0,0 +1,34 @@
+reverb_aug:
+  reverb_prob: 0.45
+  max_reverb_context: 0.5
+  rir_types: 
+    smallroom:
+      weight: 1
+      rir_path: scp:data/rirs_smallroom/rirs.scp
+      rir_norm: max
+    mediumroom:
+      weight: 1
+      rir_path: scp:data/rirs_mediumroom/rirs.scp
+      rir_norm: max
+    realroom:
+      weight: 1
+      rir_path: scp:data/rirs_real/rirs.scp
+      rir_norm: max
+noise_aug:
+  noise_prob: 0.7
+  noise_types: 
+    noise:
+      weight: 1
+      noise_path: data/musan_noise_proc_audio/wav.scp
+      min_snr: 10
+      max_snr: 20
+    music:
+      weight: 1
+      noise_path: data/musan_music_proc_audio/wav.scp
+      min_snr: 10
+      max_snr: 20
+    babble:
+      weight: 1
+      noise_path: data/musan_speech_babble/wav.scp
+      min_snr: 10
+      max_snr: 20

egs/librispeech/v0/conf/reverb_noise_aug.yaml

@@ -0,0 +1,35 @@
+reverb_aug:
+  reverb_prob: 0.45
+  max_reverb_context: 0.5
+  rir_types: 
+    smallroom:
+      weight: 1
+      rir_path: scp:data/rirs_smallroom/rirs.scp
+      rir_norm: max
+    mediumroom:
+      weight: 1
+      rir_path: scp:data/rirs_mediumroom/rirs.scp
+      rir_norm: max
+    realroom:
+      weight: 1
+      rir_path: scp:data/rirs_real/rirs.scp
+      rir_norm: max
+noise_aug:
+  noise_prob: 0.7
+  noise_types: 
+    noise:
+      weight: 1
+      noise_path: data/musan_noise_proc_audio/wav.scp
+      min_snr: 0
+      max_snr: 18
+    music:
+      weight: 1
+      noise_path: data/musan_music_proc_audio/wav.scp
+      min_snr: 3
+      max_snr: 18
+    babble:
+      weight: 1
+      noise_path: data/musan_speech_babble/wav.scp
+      min_snr: 3
+      max_snr: 18
+