CodeAnt AI: Made Changes to the file

analytics/tests/tests/trainer/test_trainer.py

@@ -11,12 +11,12 @@ class TestTrainer:
     def test_trainer_base(self):
         config = TRAINER_DEFAULTS.values()
 
-        params = dict(
+        params = {
             checkpoint_name="test",
             checkpoint_path="test_checkpoint",
             cudnn_enabled=True,
             log_dir="this/is/a/test",
-        )
+        }
         config.update(params)
         hparams = HParams(**config)
         trainer = TTSTrainer(hparams)
@@ -50,4 +50,4 @@ def test_gradient_step(self, lj_trainer):
 
         assert math.isclose(
             lj_trainer.loss[2], train_loss_4_datapoints_2_iteration, abs_tol=5e-4
-        )
+        )

uberduck_ml_dev/data/collate.py

@@ -42,7 +42,7 @@ def __call__(self, batch):
         output = Batch()
 
         if return_audio:
-            max_target_len = max([x["audio"].size(0) for x in batch])
+            max_target_len = max(x['audio'].size(0) for x in batch)
             audio_padded = torch.FloatTensor(len(batch), max_target_len)
             audio_padded.zero_()
             audio_lengths = torch.LongTensor(len(batch))
@@ -66,7 +66,7 @@ def __call__(self, batch):
 
         if return_mels:
             n_mel_channels = batch[0]["mel"].size(0)
-            max_target_len = max([x["mel"].size(1) for x in batch])
+            max_target_len = max(x['mel'].size(1) for x in batch)
             mel_padded = torch.FloatTensor(len(batch), n_mel_channels, max_target_len)
             mel_padded.zero_()
             mel_lengths = torch.LongTensor(len(batch))
@@ -90,7 +90,7 @@ def __call__(self, batch):
             output["speaker_ids"] = speaker_ids
 
         if return_f0s:
-            max_target_len = max([x["f0"].size(0) for x in batch])
+            max_target_len = max(x['f0'].size(0) for x in batch)
             f0_padded = torch.FloatTensor(len(batch), 1, max_target_len)
             f0_padded.zero_()
             for i, sample in enumerate(batch):
@@ -142,7 +142,7 @@ def __call__(self, batch):
 
         # Right zero-pad mel-spec
         num_mel_channels = batch[0]["mel"].size(0)
-        max_target_len = max([x["mel"].size(1) for x in batch])
+        max_target_len = max(x['mel'].size(1) for x in batch)
 
         # include mel padded, gate padded and speaker ids
         mel_padded = torch.FloatTensor(len(batch), num_mel_channels, max_target_len)
@@ -272,16 +272,16 @@ def __call__(self, batch):
             torch.LongTensor([x[0].size(1) for x in batch]), dim=0, descending=True
         )
 
-        max_spec_len = max([x[0].size(1) for x in batch])
-        max_wave_len = max([x[1].size(1) for x in batch])
+        max_spec_len = max(x[0].size(1) for x in batch)
+        max_wave_len = max(x[1].size(1) for x in batch)
         spec_lengths = torch.LongTensor(len(batch))
         wave_lengths = torch.LongTensor(len(batch))
         spec_padded = torch.FloatTensor(len(batch), batch[0][0].size(0), max_spec_len)
         wave_padded = torch.FloatTensor(len(batch), 1, max_wave_len)
         spec_padded.zero_()
         wave_padded.zero_()
 
-        max_phone_len = max([x[2].size(0) for x in batch])
+        max_phone_len = max(x[2].size(0) for x in batch)
         phone_lengths = torch.LongTensor(len(batch))
         phone_padded = torch.FloatTensor(
             len(batch), max_phone_len, batch[0][2].shape[1]
@@ -325,4 +325,4 @@ def __call__(self, batch):
             wave_padded,
             wave_lengths,
             sid,  # nee "dv"
-        )
+        )

uberduck_ml_dev/data/data.py

@@ -16,29 +16,18 @@
 from librosa import pyin
 import parselmouth
 
-from ..models.common import MelSTFT
+from ..models.common import FILTER_LENGTH, HOP_LENGTH, MelSTFT, N_MEL_CHANNELS, SAMPLING_RATE, TacotronSTFT, WIN_LENGTH
 from ..utils.utils import (
     load_filepaths_and_text,
     intersperse,
 )
-from .utils import (
-    oversample,
-    _orig_to_dense_speaker_id,
-    beta_binomial_prior_distribution,
-)
+from .utils import _orig_to_dense_speaker_id, beta_binomial_prior_distribution, oversample, spectrogram_torch
 from ..text.utils import text_to_sequence
 from ..text.text_processing import TextProcessing
-from ..models.common import (
-    FILTER_LENGTH,
-    HOP_LENGTH,
-    WIN_LENGTH,
-    SAMPLING_RATE,
-    N_MEL_CHANNELS,
-    TacotronSTFT,
-)
+
 from ..text.symbols import NVIDIA_TACO2_SYMBOLS
 from ..trainer.rvc.utils import load_wav_to_torch
-from .utils import spectrogram_torch
+
 
 
 from ..models.tacotron2 import MAX_WAV_VALUE
@@ -185,7 +174,7 @@ def __init__(
     # NOTE (Sam): this is the RADTTS version - more recent than mellotron from the same author.
     # NOTE (Sam): in contrast to get_gst, the computation here is kept in this file rather than a functional argument.
     def _get_f0(self, audiopath, audio):
-        filename = "_".join(audiopath.split("/")).split(".wav")[0]
+        filename = '_'.join(audiopath.split('/')).split('.wav', maxsplit=1)[0]
         f0_path = os.path.join(self.f0_cache_path, filename)
         f0_path += "_f0_sr{}_fl{}_hl{}_f0min{}_f0max{}_log{}.pt".format(
             self.sampling_rate,
@@ -337,10 +326,7 @@ def __len__(self):
 
     def sample_test_batch(self, size):
         idx = np.random.choice(range(len(self)), size=size, replace=False)
-        test_batch = []
-        for index in idx:
-            test_batch.append(self.__getitem__(index))
-        return test_batch
+        return [self.__getitem__(index) for index in idx]
 
     def get_f0_pvoiced(
         self,
@@ -505,6 +491,8 @@ def __init__(
 
     def load_data(self, datasets, split="|"):
         dataset = []
+
+        duration = -1
         for dset_name, dset_dict in datasets.items():
             folder_path = dset_dict["basedir"]
             audiodir = dset_dict["audiodir"]
@@ -521,9 +509,11 @@ def load_data(self, datasets, split="|"):
             with open(filelist_path, encoding="utf-8") as f:
                 data = [line.strip().split(split) for line in f]
 
+
+
             for d in data:
                 # NOTE (Sam): BEWARE! change/comment depending on filelist.
-                duration = -1
+
                 dataset.append(
                     {
                         "audiopath": os.path.join(wav_folder_prefix, d[0]),
@@ -776,9 +766,7 @@ def __getitem__(self, idx):
         return None
 
     def __len__(self):
-        nfiles = len(self.audiopaths)
-
-        return nfiles
+        return len(self.audiopaths)
 
 
 def get_f0_pvoiced(
@@ -968,9 +956,7 @@ def __getitem__(self, idx):
         return None
 
     def __len__(self):
-        nfiles = len(self.audiopaths)
-
-        return nfiles
+        return len(self.audiopaths)
 
 
 RADTTS_DEFAULTS = {
@@ -1073,8 +1059,7 @@ def _filter(self):
         self.lengths = lengths
 
     def get_sid(self, sid):
-        sid = torch.LongTensor([int(sid)])
-        return sid
+        return torch.LongTensor([int(sid)])
 
     def get_audio_text_pair(self, audiopath_and_text):
         file = audiopath_and_text[0]
@@ -1194,9 +1179,11 @@ def _create_buckets(self):
                 self.boundaries.pop(i + 1)
 
         num_samples_per_bucket = []
+
+        total_batch_size = self.num_replicas * self.batch_size
         for i in range(len(buckets)):
             len_bucket = len(buckets[i])
-            total_batch_size = self.num_replicas * self.batch_size
+
             rem = (
                 total_batch_size - (len_bucket % total_batch_size)
             ) % total_batch_size
@@ -1256,17 +1243,16 @@ def _bisect(self, x, lo=0, hi=None):
         if hi is None:
             hi = len(self.boundaries) - 1
 
-        if hi > lo:
-            mid = (hi + lo) // 2
-            if self.boundaries[mid] < x and x <= self.boundaries[mid + 1]:
-                return mid
-            elif x <= self.boundaries[mid]:
-                return self._bisect(x, lo, mid)
-            else:
-                return self._bisect(x, mid + 1, hi)
-        else:
+        if hi <= lo:
             return -1
 
+        mid = (hi + lo) // 2
+        if self.boundaries[mid] < x and x <= self.boundaries[mid + 1]:
+            return mid
+        if x <= self.boundaries[mid]:
+            return self._bisect(x, lo, mid)
+        return self._bisect(x, mid + 1, hi)
+
     def __len__(self):
         return self.num_samples // self.batch_size
 
@@ -1331,4 +1317,4 @@ def __getitem__(self, index):
         return output
 
     def __len__(self):
-        return len(self.data_paths)
+        return len(self.data_paths)