fix other recipes

speechbrain · Jun 17, 2022 · a812c9b · a812c9b
1 parent 507b69b
commit a812c9b
Show file tree

Hide file tree

Showing 18 changed files with 114 additions and 48 deletions.
diff --git a/recipes/IEMOCAP/emotion_recognition/train.py b/recipes/IEMOCAP/emotion_recognition/train.py
@@ -269,9 +269,14 @@ def label_pipeline(emo):
     # Define datasets. We also connect the dataset with the data processing
     # functions defined above.
     datasets = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[audio_pipeline, label_pipeline],
             output_keys=["id", "sig", "emo_encoded"],

diff --git a/recipes/IEMOCAP/emotion_recognition/train_with_wav2vec2.py b/recipes/IEMOCAP/emotion_recognition/train_with_wav2vec2.py
@@ -195,9 +195,14 @@ def label_pipeline(emo):
     # Define datasets. We also connect the dataset with the data processing
     # functions defined above.
     datasets = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[audio_pipeline, label_pipeline],
             output_keys=["id", "sig", "emo_encoded"],

diff --git a/recipes/UrbanSound8k/SoundClassification/train.py b/recipes/UrbanSound8k/SoundClassification/train.py
@@ -154,7 +154,7 @@ def on_stage_start(self, stage, epoch=None):
         """
         # Set up statistics trackers for this stage
         self.loss_metric = sb.utils.metric_stats.MetricStats(
-            metric=sb.nnet.losses.nll_loss  # TODO put in yaml hparams?
+            metric=sb.nnet.losses.nll_loss
         )
 
         # Compute Accuracy using MetricStats
@@ -347,9 +347,14 @@ def label_pipeline(class_string):
     # Define datasets. We also connect the dataset with the data processing
     # functions defined above.
     datasets = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[audio_pipeline, label_pipeline],
             output_keys=["id", "sig", "class_string_encoded"],

diff --git a/recipes/Voicebank/ASR/CTC/train.py b/recipes/Voicebank/ASR/CTC/train.py
@@ -118,9 +118,14 @@ def text_pipeline(phones):
 
     # 3. Create datasets
     data = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         data[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[audio_pipeline, text_pipeline],
             output_keys=["id", "sig", "phn_encoded"],

diff --git a/recipes/Voicebank/MTL/ASR_enhance/train.py b/recipes/Voicebank/MTL/ASR_enhance/train.py
@@ -385,9 +385,14 @@ def target_pipeline(target):
 
     # Create datasets
     data = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         data[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[audio_pipeline, target_pipeline],
             output_keys=["id", "noisy_sig", "clean_sig"] + token_keys,

diff --git a/recipes/Voicebank/MTL/CoopNet/train.py b/recipes/Voicebank/MTL/CoopNet/train.py
@@ -530,9 +530,14 @@ def target_pipeline(target):
 
     # 4. Create datasets
     data = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         data[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[noisy_pipeline, clean_pipeline, target_pipeline],
             output_keys=["id", "noisy_sig", "clean_sig", "phn_encoded"],

diff --git a/recipes/Voicebank/dereverb/MetricGAN-U/train.py b/recipes/Voicebank/dereverb/MetricGAN-U/train.py
@@ -678,21 +678,24 @@ def dataio_prep(hparams):
 
     # Define datasets
     datasets = {}
-
-    dataset = "train"
-    datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-        json_path=hparams[f"{dataset}_annotation"],
+    datasets["train"] = sb.dataio.dataset.DynamicItemDataset.from_json(
+        json_path=hparams["train_annotation"],
         replacements={"data_root": hparams["data_folder"]},
         dynamic_items=[audio_pipeline_train],
         output_keys=["id", "noisy_sig"],
     )
-    for dataset in ["valid", "test"]:
-        datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
-            replacements={"data_root": hparams["data_folder"]},
-            dynamic_items=[audio_pipeline_valid],
-            output_keys=["id", "noisy_sig", "clean_sig"],
-        )
+    datasets["valid"] = sb.dataio.dataset.DynamicItemDataset.from_json(
+        json_path=hparams["valid_annotation"],
+        replacements={"data_root": hparams["data_folder"]},
+        dynamic_items=[audio_pipeline_valid],
+        output_keys=["id", "noisy_sig", "clean_sig"],
+    )
+    datasets["test"] = sb.dataio.dataset.DynamicItemDataset.from_json(
+        json_path=hparams["test_annotation"],
+        replacements={"data_root": hparams["data_folder"]},
+        dynamic_items=[audio_pipeline_valid],
+        output_keys=["id", "noisy_sig", "clean_sig"],
+    )
 
     return datasets
 

diff --git a/recipes/Voicebank/dereverb/spectral_mask/hparams/train.yaml b/recipes/Voicebank/dereverb/spectral_mask/hparams/train.yaml
@@ -30,9 +30,9 @@ N_fft: 512
 window_fn: !name:torch.hamming_window
 
 # Data files
-train_annotation: !ref <data_folder>/train_revb.json
-valid_annotation: !ref <data_folder>/valid_revb.json
-test_annotation: !ref <data_folder>/test_revb.json
+train_annotation: !ref <output_folder>/train_revb.json
+valid_annotation: !ref <output_folder>/valid_revb.json
+test_annotation: !ref <output_folder>/test_revb.json
 skip_prep: False
 
 # Training Parameters

diff --git a/recipes/Voicebank/dereverb/spectral_mask/train.py b/recipes/Voicebank/dereverb/spectral_mask/train.py
@@ -167,9 +167,14 @@ def clean_pipeline(clean_wav):
 
     # Define datasets
     datasets = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[noisy_pipeline, clean_pipeline],
             output_keys=["id", "noisy_sig", "clean_sig"],

diff --git a/recipes/Voicebank/dereverb/spectral_mask/voicebank_revb_prepare.py b/recipes/Voicebank/dereverb/spectral_mask/voicebank_revb_prepare.py
@@ -175,7 +175,6 @@ def prepare_voicebank(
     >>> save_folder = 'exp/Voicebank_exp'
     >>> prepare_voicebank(data_folder, save_folder)
     """
-
     if skip_prep:
         return
 

diff --git a/recipes/Voicebank/enhance/MetricGAN-U/README.md b/recipes/Voicebank/enhance/MetricGAN-U/README.md
@@ -9,6 +9,7 @@ Use the `download_vctk` function in `voicebank_prepare.py` to download the datas
 and resample it to 16000 Hz. To run an experiment, execute the following command in
 the current folder:
 
+
 ```bash
 python train.py hparams/train_dnsmos.yaml --data_folder /path/to/data_folder
 ```

diff --git a/recipes/Voicebank/enhance/MetricGAN-U/extra-requirements.txt b/recipes/Voicebank/enhance/MetricGAN-U/extra-requirements.txt
@@ -0,0 +1 @@
+git+https://github.com/jfsantos/SRMRpy
diff --git a/recipes/Voicebank/enhance/MetricGAN-U/train.py b/recipes/Voicebank/enhance/MetricGAN-U/train.py
@@ -669,20 +669,25 @@ def dataio_prep(hparams):
     # Define datasets
     datasets = {}
 
-    dataset = "train"
-    datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-        json_path=hparams[f"{dataset}_annotation"],
+    datasets["train"] = sb.dataio.dataset.DynamicItemDataset.from_json(
+        json_path=hparams["train_annotation"],
         replacements={"data_root": hparams["data_folder"]},
         dynamic_items=[audio_pipeline_train],
         output_keys=["id", "noisy_sig"],
     )
-    for dataset in ["valid", "test"]:
-        datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
-            replacements={"data_root": hparams["data_folder"]},
-            dynamic_items=[audio_pipeline_valid],
-            output_keys=["id", "noisy_sig", "clean_sig"],
-        )
+    datasets["valid"] = sb.dataio.dataset.DynamicItemDataset.from_json(
+        json_path=hparams["valid_annotation"],
+        replacements={"data_root": hparams["data_folder"]},
+        dynamic_items=[audio_pipeline_valid],
+        output_keys=["id", "noisy_sig", "clean_sig"],
+    )
+
+    datasets["test"] = sb.dataio.dataset.DynamicItemDataset.from_json(
+        json_path=hparams["test_annotation"],
+        replacements={"data_root": hparams["data_folder"]},
+        dynamic_items=[audio_pipeline_valid],
+        output_keys=["id", "noisy_sig", "clean_sig"],
+    )
 
     return datasets
 

diff --git a/recipes/Voicebank/enhance/MetricGAN/extra_requirements.txt b/recipes/Voicebank/enhance/MetricGAN/extra_requirements.txt
@@ -0,0 +1,2 @@
+pesq
+
diff --git a/recipes/Voicebank/enhance/MetricGAN/train.py b/recipes/Voicebank/enhance/MetricGAN/train.py
@@ -511,9 +511,14 @@ def dataio_prep(hparams):
 
     # Define datasets
     datasets = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[audio_pipeline],
             output_keys=["id", "noisy_sig", "clean_sig"],

diff --git a/recipes/Voicebank/enhance/SEGAN/train.py b/recipes/Voicebank/enhance/SEGAN/train.py
@@ -409,9 +409,14 @@ def clean_pipeline(clean_wav):
 
     # Define datasets
     datasets = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[noisy_pipeline, clean_pipeline],
             output_keys=["id", "noisy_sig", "clean_sig"],

diff --git a/recipes/Voicebank/enhance/spectral_mask/train.py b/recipes/Voicebank/enhance/spectral_mask/train.py
@@ -163,9 +163,14 @@ def clean_pipeline(clean_wav):
 
     # Define datasets
     datasets = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[noisy_pipeline, clean_pipeline],
             output_keys=["id", "noisy_sig", "clean_sig"],

diff --git a/recipes/Voicebank/enhance/waveform_map/train.py b/recipes/Voicebank/enhance/waveform_map/train.py
@@ -140,9 +140,14 @@ def clean_pipeline(clean_wav):
 
     # Define datasets
     datasets = {}
-    for dataset in ["train", "valid", "test"]:
+    data_info = {
+        "train": hparams["train_annotation"],
+        "valid": hparams["valid_annotation"],
+        "test": hparams["test_annotation"],
+    }
+    for dataset in data_info:
         datasets[dataset] = sb.dataio.dataset.DynamicItemDataset.from_json(
-            json_path=hparams[f"{dataset}_annotation"],
+            json_path=data_info[dataset],
             replacements={"data_root": hparams["data_folder"]},
             dynamic_items=[noisy_pipeline, clean_pipeline],
             output_keys=["id", "noisy_sig", "clean_sig"],
-Original file line number
+Diff line change
@@ Expand Up / @@ -175,7 +175,6 @@ def prepare_voicebank( @@
         >>> save_folder = 'exp/Voicebank_exp'
         >>> prepare_voicebank(data_folder, save_folder)
         """
         if skip_prep:
             return
@@ Expand Down @@