Initial commit

Author: bshall

.gitignore

@@ -110,6 +110,9 @@ ENV/
 env.bak/
 venv.bak/
 
+# PyCharm project settings
+.idea
+
 # Spyder project settings
 .spyderproject
 .spyproject
@@ -127,3 +130,6 @@ dmypy.json
 
 # Pyre type checker
 .pyre/
+
+# Data files
+.npy

config.json

@@ -0,0 +1,29 @@
+{
+  "preprocessing": {
+    "sample_rate": 16000,
+    "num_fft": 2048,
+    "num_mels": 80,
+    "fmin": 50,
+    "preemph": 0.97,
+    "top_db": 80,
+    "hop_length": 160,
+    "win_length": 400
+  },
+
+  "model": {
+    "encoder_channels": 256,
+    "z_dim": 256,
+    "c_dim": 256
+  },
+
+  "training": {
+    "n_sample_frames": 128,
+    "n_speakers": 8,
+    "n_utterances_per_speaker": 6,
+    "learning_rate": 2e-4,
+    "n_prediction_steps": 12,
+    "n_negatives": 17,
+    "n_epochs": 5000,
+    "checkpoint_interval": 500
+  }
+}

dataset.py

@@ -0,0 +1,46 @@
+import numpy as np
+import torch
+import csv
+import random
+from torch.utils.data import Dataset
+from pathlib import Path
+
+
+class SpeakerDataset(Dataset):
+    def __init__(self, root, n_sample_frames, n_utterances_per_speaker, hop_length, sample_rate):
+        self.root = Path(root)
+        self.n_sample_frames = n_sample_frames
+        self.n_utterances_per_speaker = n_utterances_per_speaker
+
+        with open(self.root / "speakers.csv") as file:
+            reader = csv.reader(file)
+            self.speakers = sorted([speaker for speaker, in reader])
+
+        min_duration = n_sample_frames * hop_length / sample_rate
+        with open(self.root / "train.csv") as file:
+            reader = csv.reader(file)
+            metadata = dict()
+            for _, _, duration, out_path in reader:
+                if float(duration) > min_duration:
+                    out_path = Path(out_path)
+                    speaker = out_path.stem.split("_")[0]
+                    metadata.setdefault(speaker, []).append(out_path)
+            self.metadata = [(k, v) for k, v in metadata.items() if len(v) >= n_utterances_per_speaker]
+
+    def __len__(self):
+        return len(self.metadata)
+
+    def __getitem__(self, index):
+        speaker, paths = self.metadata[index]
+
+        mels = list()
+        paths = random.sample(paths, self.n_utterances_per_speaker)
+        for path in paths:
+            path = self.root.parent / path
+            mel = np.load(path.with_suffix(".mel.npy"))
+            pos = random.randint(0, mel.shape[1] - self.n_sample_frames)
+            mel = mel[:, pos:pos + self.n_sample_frames]
+            mels.append(mel)
+        mels = np.stack(mels)
+
+        return torch.from_numpy(mels), self.speakers.index(speaker)

datasets/ZeroSpeech2019/english/speakers.csv

@@ -0,0 +1,102 @@
+S015
+S020
+S021
+S023
+S027
+S031
+S032
+S033
+S034
+S035
+S036
+S037
+S038
+S039
+S040
+S041
+S042
+S043
+S044
+S045
+S046
+S047
+S048
+S049
+S050
+S051
+S052
+S053
+S054
+S055
+S056
+S058
+S059
+S060
+S061
+S063
+S064
+S065
+S066
+S067
+S069
+S070
+S071
+S072
+S073
+S074
+S075
+S076
+S077
+S078
+S079
+S080
+S082
+S083
+S084
+S085
+S086
+S087
+S088
+S090
+S091
+S092
+S093
+S094
+S095
+S096
+S097
+S098
+S099
+S100
+S101
+S102
+S103
+S104
+S105
+S106
+S107
+S109
+S110
+S111
+S112
+S113
+S114
+S115
+S116
+S117
+S118
+S119
+S120
+S121
+S122
+S123
+S125
+S126
+S127
+S128
+S129
+S131
+S132
+S133
+V001
+V002

datasets/ZeroSpeech2019/english/synthesis.csv

@@ -0,0 +1,193 @@
+english/test/S002_0379088085,V002
+english/test/S002_3329644604,V002
+english/test/S002_0716659834,V002
+english/test/S002_0034198090,V001
+english/test/S002_3675593510,V001
+english/test/S002_1248863180,V002
+english/test/S002_1811315103,V001
+english/test/S002_0150793611,V001
+english/test/S002_0488457066,V002
+english/test/S002_1927813378,V002
+english/test/S002_3835497764,V001
+english/test/S002_2613999357,V001
+english/test/S002_3397632494,V001
+english/test/S003_2650563980,V001
+english/test/S003_3628811388,V001
+english/test/S003_2246977596,V001
+english/test/S003_4143021831,V002
+english/test/S003_2221495694,V002
+english/test/S003_1178890909,V002
+english/test/S003_2611431933,V002
+english/test/S003_2240019245,V001
+english/test/S003_2299041764,V002
+english/test/S003_1478639117,V001
+english/test/S003_3732887959,V001
+english/test/S003_0348166629,V002
+english/test/S003_2098261986,V001
+english/test/S005_1869043913,V001
+english/test/S005_2462667879,V002
+english/test/S005_2857012128,V002
+english/test/S005_0337854417,V001
+english/test/S005_1255051908,V001
+english/test/S005_3817578655,V001
+english/test/S005_1369359729,V002
+english/test/S005_2339421144,V002
+english/test/S005_1674880857,V001
+english/test/S005_1027254890,V001
+english/test/S005_0525058276,V002
+english/test/S005_3335289900,V001
+english/test/S005_2243904656,V002
+english/test/S005_3148422380,V002
+english/test/S006_3547507286,V001
+english/test/S006_0256405871,V001
+english/test/S006_2402232073,V001
+english/test/S006_1744073907,V002
+english/test/S006_0791080940,V002
+english/test/S006_2309191462,V002
+english/test/S006_2467127665,V002
+english/test/S006_3907468490,V001
+english/test/S006_1919833715,V002
+english/test/S006_2233331201,V001
+english/test/S006_2068766372,V001
+english/test/S006_1788089822,V002
+english/test/S006_1876582974,V001
+english/test/S007_1670529004,V001
+english/test/S007_2106426595,V001
+english/test/S007_1328762623,V002
+english/test/S007_0678239348,V001
+english/test/S007_3243217924,V002
+english/test/S007_0204997433,V001
+english/test/S007_3771169705,V002
+english/test/S007_2836644370,V001
+english/test/S007_1179982703,V002
+english/test/S008_4096000418,V001
+english/test/S008_0374594724,V001
+english/test/S008_0316936606,V002
+english/test/S008_2311158062,V001
+english/test/S008_2210326791,V002
+english/test/S008_4101672507,V001
+english/test/S008_1533376143,V002
+english/test/S008_2684330882,V001
+english/test/S008_1680359331,V002
+english/test/S009_1134572964,V001
+english/test/S009_0951978018,V001
+english/test/S009_2963764176,V002
+english/test/S009_3172133024,V002
+english/test/S009_2409126958,V001
+english/test/S009_2230685767,V001
+english/test/S009_1637278923,V002
+english/test/S009_2554341485,V001
+english/test/S009_2664556587,V001
+english/test/S009_3009558539,V002
+english/test/S009_2922384978,V002
+english/test/S009_3769715709,V002
+english/test/S009_3852218745,V002
+english/test/S009_0722532299,V001
+english/test/S009_2675987928,V001
+english/test/S011_3385425823,V001
+english/test/S011_1485461930,V002
+english/test/S011_4204160922,V002
+english/test/S011_1220140001,V001
+english/test/S011_1056847685,V001
+english/test/S011_4019527684,V001
+english/test/S011_1004261560,V002
+english/test/S011_2504494747,V002
+english/test/S011_2020332133,V001
+english/test/S011_0826602971,V001
+english/test/S011_4248927546,V002
+english/test/S011_0538842874,V001
+english/test/S011_1486905221,V002
+english/test/S011_3070188133,V002
+english/test/S013_1647107851,V001
+english/test/S013_2857211736,V001
+english/test/S013_1546167225,V001
+english/test/S013_0990512588,V002
+english/test/S013_3906392480,V002
+english/test/S013_0694888237,V002
+english/test/S013_1319297801,V002
+english/test/S013_0717685577,V001
+english/test/S013_3843497491,V002
+english/test/S013_0461257464,V001
+english/test/S013_2195076294,V001
+english/test/S013_1720378819,V002
+english/test/S013_3050055609,V001
+english/test/S016_0033179769,V001
+english/test/S016_4001194935,V002
+english/test/S016_2410213271,V002
+english/test/S016_3036296225,V001
+english/test/S016_0909866182,V001
+english/test/S016_1604362137,V001
+english/test/S016_4241276147,V002
+english/test/S016_4079864915,V002
+english/test/S016_1517581208,V001
+english/test/S016_1426116964,V001
+english/test/S016_0015229803,V002
+english/test/S016_2853428649,V001
+english/test/S016_0007211729,V002
+english/test/S016_4184772113,V002
+english/test/S019_2459060426,V002
+english/test/S019_1203011745,V002
+english/test/S019_2784269462,V002
+english/test/S019_1370348056,V001
+english/test/S019_2465622735,V001
+english/test/S019_3655501544,V002
+english/test/S019_0444395040,V001
+english/test/S019_1864848970,V001
+english/test/S019_2946419609,V002
+english/test/S019_0163264856,V002
+english/test/S019_3566025888,V001
+english/test/S019_0876437283,V001
+english/test/S019_1517717992,V001
+english/test/S022_4213163508,V001
+english/test/S022_1587635681,V001
+english/test/S022_1579158408,V002
+english/test/S022_1376425751,V002
+english/test/S022_2936194818,V001
+english/test/S022_1341990990,V001
+english/test/S022_3801674752,V002
+english/test/S022_2436309351,V001
+english/test/S022_2467872015,V001
+english/test/S022_3587271288,V002
+english/test/S022_1990728118,V002
+english/test/S022_3599507778,V002
+english/test/S022_0598465739,V002
+english/test/S022_0799854662,V001
+english/test/S022_2211405183,V001
+english/test/S029_0296750778,V001
+english/test/S029_4256753032,V001
+english/test/S029_4193376528,V001
+english/test/S029_2159765033,V002
+english/test/S029_4188495052,V002
+english/test/S029_0319474799,V001
+english/test/S029_1929874338,V001
+english/test/S029_2318301126,V001
+english/test/S029_0016520932,V001
+english/test/S029_1068142416,V002
+english/test/S029_4050153366,V001
+english/test/S029_0601534788,V001
+english/test/S029_1148954548,V002
+english/test/S029_2285517287,V002
+english/test/S029_1480335310,V001
+english/test/S029_3513270162,V001
+english/test/S029_3694116817,V001
+english/test/S029_1644893934,V001
+english/test/S030_1756493637,V002
+english/test/S030_3313878040,V002
+english/test/S030_2768730427,V002
+english/test/S030_0405736610,V001
+english/test/S030_3045276808,V001
+english/test/S030_1317267807,V002
+english/test/S030_2540374847,V002
+english/test/S030_2477652858,V002
+english/test/S030_1644595354,V002
+english/test/S030_0038139088,V001
+english/test/S030_1515780978,V001
+english/test/S030_1930457209,V002
+english/test/S030_3624094171,V001
+english/test/S030_1524414356,V001
+english/test/S030_2091192866,V001
+english/test/S030_0730141577,V001
+english/test/S030_3923466856,V001
+english/test/S030_2739033792,V001
+english/test/S030_0244072828,V001
+english/test/S030_0378256235,V001