black fix

egs2/mini_librispeech/diar1/local/simulation/make_mixture.py

@@ -106,8 +106,8 @@
     else:
         noise_data = noise_data[:maxlen]
     # noise power is scaled according to selected SNR, then mixed
-    signal_power = np.sum(mixture ** 2) / len(mixture)
-    noise_power = np.sum(noise_data ** 2) / len(noise_data)
+    signal_power = np.sum(mixture**2) / len(mixture)
+    noise_power = np.sum(noise_data**2) / len(noise_data)
     scale = math.sqrt(math.pow(10, -noise_snr / 10) * signal_power / noise_power)
     mixture += noise_data * scale
     # output the wav file and write wav.scp

egs2/mini_librispeech/diar1/local/simulation/make_mixture_nooverlap.py

@@ -90,8 +90,8 @@
     else:
         noise_data = noise_data[:maxlen]
     # noise power is scaled according to selected SNR, then mixed
-    signal_power = np.sum(mixture ** 2) / len(mixture)
-    noise_power = np.sum(noise_data ** 2) / len(noise_data)
+    signal_power = np.sum(mixture**2) / len(mixture)
+    noise_power = np.sum(noise_data**2) / len(noise_data)
     scale = math.sqrt(math.pow(10, -noise_snr / 10) * signal_power / noise_power)
     mixture += noise_data * scale
     # output the wav file and write wav.scp

espnet/asr/asr_utils.py

@@ -666,7 +666,7 @@ def add_gradient_noise(model, iteration, duration=100, eta=1.0, scale_factor=0.5
         scale_factor (float) {0.55}: The scale of `sigma`.
     """
     interval = (iteration // duration) + 1
-    sigma = eta / interval ** scale_factor
+    sigma = eta / interval**scale_factor
     for param in model.parameters():
         if param.grad is not None:
             _shape = param.grad.size()

espnet/nets/chainer_backend/e2e_asr_transformer.py

@@ -154,7 +154,7 @@ def __init__(self, idim, odim, args, ignore_id=-1, flag_return=True):
         self.char_list = args.char_list
         self.space = args.sym_space
         self.blank = args.sym_blank
-        self.scale_emb = args.adim ** 0.5
+        self.scale_emb = args.adim**0.5
         self.sos = odim - 1
         self.eos = odim - 1
         self.subsample = get_subsample(args, mode="asr", arch="transformer")

espnet/nets/pytorch_backend/e2e_asr_mix.py

@@ -323,7 +323,7 @@ def forward(self, xs_pad, ilens, ys_pad):
                             hlens[i // self.num_spkrs],
                             ys_pad[i % self.num_spkrs],
                         )
-                        for i in range(self.num_spkrs ** 2)
+                        for i in range(self.num_spkrs**2)
                     ],
                     dim=1,
                 )  # (B, num_spkrs^2)
@@ -441,13 +441,13 @@ def forward(self, xs_pad, ilens, ys_pad):
                     editdistance.eval(
                         hyp_words[ns // self.num_spkrs], ref_words[ns % self.num_spkrs]
                     )
-                    for ns in range(self.num_spkrs ** 2)
+                    for ns in range(self.num_spkrs**2)
                 ]  # h1r1,h1r2,h2r1,h2r2
                 tmp_char_ed = [
                     editdistance.eval(
                         hyp_chars[ns // self.num_spkrs], ref_chars[ns % self.num_spkrs]
                     )
-                    for ns in range(self.num_spkrs ** 2)
+                    for ns in range(self.num_spkrs**2)
                 ]  # h1r1,h1r2,h2r1,h2r2
 
                 word_eds.append(self.pit.min_pit_sample(torch.tensor(tmp_word_ed))[0])
@@ -676,7 +676,7 @@ def calculate_all_attentions(self, xs_pad, ilens, ys_pad):
                             hlens[i // self.num_spkrs],
                             ys_pad[i % self.num_spkrs],
                         )
-                        for i in range(self.num_spkrs ** 2)
+                        for i in range(self.num_spkrs**2)
                     ],
                     1,
                 )  # (B, num_spkrs^2)

espnet/nets/pytorch_backend/e2e_asr_mix_transformer.py

@@ -124,7 +124,7 @@ def forward(self, xs_pad, ilens, ys_pad):
                     hs_len[i // self.num_spkrs],
                     ys_pad[i % self.num_spkrs],
                 )
-                for i in range(self.num_spkrs ** 2)
+                for i in range(self.num_spkrs**2)
             ],
             dim=1,
         )  # (B, num_spkrs^2)

espnet/nets/pytorch_backend/e2e_mt_transformer.py

@@ -141,11 +141,11 @@ def reset_parameters(self, args):
         """Initialize parameters."""
         initialize(self, args.transformer_init)
         torch.nn.init.normal_(
-            self.encoder.embed[0].weight, mean=0, std=args.adim ** -0.5
+            self.encoder.embed[0].weight, mean=0, std=args.adim**-0.5
         )
         torch.nn.init.constant_(self.encoder.embed[0].weight[self.pad], 0)
         torch.nn.init.normal_(
-            self.decoder.embed[0].weight, mean=0, std=args.adim ** -0.5
+            self.decoder.embed[0].weight, mean=0, std=args.adim**-0.5
         )
         torch.nn.init.constant_(self.decoder.embed[0].weight[self.pad], 0)
 

espnet/nets/pytorch_backend/e2e_st_transformer.py

@@ -185,7 +185,7 @@ def reset_parameters(self, args):
         initialize(self, args.transformer_init)
         if self.mt_weight > 0:
             torch.nn.init.normal_(
-                self.encoder_mt.embed[0].weight, mean=0, std=args.adim ** -0.5
+                self.encoder_mt.embed[0].weight, mean=0, std=args.adim**-0.5
             )
             torch.nn.init.constant_(self.encoder_mt.embed[0].weight[self.pad], 0)
 

espnet/nets/pytorch_backend/e2e_tts_tacotron2.py

@@ -121,7 +121,7 @@ def _make_guided_attention_mask(ilen, olen, sigma):
         grid_x, grid_y = torch.meshgrid(torch.arange(olen), torch.arange(ilen))
         grid_x, grid_y = grid_x.float().to(olen.device), grid_y.float().to(ilen.device)
         return 1.0 - torch.exp(
-            -((grid_y / ilen - grid_x / olen) ** 2) / (2 * (sigma ** 2))
+            -((grid_y / ilen - grid_x / olen) ** 2) / (2 * (sigma**2))
         )
 
     @staticmethod

espnet/nets/pytorch_backend/frontends/dnn_beamformer.py

@@ -162,7 +162,7 @@ def forward(
         psd = (psd.sum(dim=-1) / (C - 1)).transpose(-1, -2)
 
         # Calculate amplitude
-        psd_feat = (psd.real ** 2 + psd.imag ** 2) ** 0.5
+        psd_feat = (psd.real**2 + psd.imag**2) ** 0.5
 
         # (B, C, F) -> (B, C, F2)
         mlp_psd = self.mlp_psd(psd_feat)

espnet/nets/pytorch_backend/frontends/dnn_wpe.py

@@ -62,7 +62,7 @@ def forward(
 
         for i in range(self.iterations):
             # Calculate power: (..., C, T)
-            power = enhanced.real ** 2 + enhanced.imag ** 2
+            power = enhanced.real**2 + enhanced.imag**2
             if i == 0 and self.use_dnn_mask:
                 # mask: (B, F, C, T)
                 (mask,), _ = self.mask_est(enhanced, ilens)

espnet/nets/pytorch_backend/frontends/feature_transform.py

@@ -64,7 +64,7 @@ def forward(
             h = x
 
         # h: ComplexTensor(B, T, F) -> torch.Tensor(B, T, F)
-        h = h.real ** 2 + h.imag ** 2
+        h = h.real**2 + h.imag**2
 
         h, _ = self.logmel(h, ilens)
         if self.stats_file is not None:

espnet/nets/pytorch_backend/frontends/mask_estimator.py

@@ -46,7 +46,7 @@ def forward(
         xs = xs.permute(0, 2, 3, 1)
 
         # Calculate amplitude: (B, C, T, F) -> (B, C, T, F)
-        xs = (xs.real ** 2 + xs.imag ** 2) ** 0.5
+        xs = (xs.real**2 + xs.imag**2) ** 0.5
         # xs: (B, C, T, F) -> xs: (B * C, T, F)
         xs = xs.contiguous().view(-1, xs.size(-2), xs.size(-1))
         # ilens: (B,) -> ilens_: (B * C)

espnet/nets/pytorch_backend/wavenet.py

@@ -202,7 +202,7 @@ def __init__(
         self.upsampling_factor = upsampling_factor
 
         self.dilations = [
-            2 ** i for i in range(self.dilation_depth)
+            2**i for i in range(self.dilation_depth)
         ] * self.dilation_repeat
         self.receptive_field = (self.kernel_size - 1) * sum(self.dilations) + 1
 

espnet/scheduler/scheduler.py

@@ -135,12 +135,12 @@ def _add_arguments(parser: _PrefixParser):
     def __init__(self, key, args):
         """Initialize class."""
         super().__init__(key, args)
-        self.normalize = 1 / (self.warmup * self.warmup ** -1.5)
+        self.normalize = 1 / (self.warmup * self.warmup**-1.5)
 
     def scale(self, step):
         """Scale of lr."""
         step += 1  # because step starts from 0
-        return self.normalize * min(step ** -0.5, step * self.warmup ** -1.5)
+        return self.normalize * min(step**-0.5, step * self.warmup**-1.5)
 
 
 @register_scheduler

espnet/transform/add_deltas.py

@@ -9,7 +9,7 @@ def delta(feat, window):
         delta_feat[i:] += -i * feat[:-i]
         delta_feat[-i:] += i * feat[-1]
         delta_feat[:i] += -i * feat[0]
-    delta_feat /= 2 * sum(i ** 2 for i in range(1, window + 1))
+    delta_feat /= 2 * sum(i**2 for i in range(1, window + 1))
     return delta_feat
 
 

espnet/transform/cmvn.py

@@ -130,14 +130,14 @@ def __repr__(self):
 
     def __call__(self, x, uttid=None):
         # x: [Time, Dim]
-        square_sums = (x ** 2).sum(axis=0)
+        square_sums = (x**2).sum(axis=0)
         mean = x.mean(axis=0)
 
         if self.norm_means:
             x = np.subtract(x, mean)
 
         if self.norm_vars:
-            var = square_sums / x.shape[0] - mean ** 2
+            var = square_sums / x.shape[0] - mean**2
             std = np.maximum(np.sqrt(var), self.std_floor)
             x = np.divide(x, std)
 

espnet/transform/perturb.py

@@ -270,7 +270,7 @@ def __call__(self, x, uttid=None, train=True):
 
         if self.dbunit:
             ratio = 10 ** (ratio / 20)
-        scale = ratio * numpy.sqrt((x ** 2).mean())
+        scale = ratio * numpy.sqrt((x**2).mean())
 
         # 2. Get noise
         if self.utt2noise is not None:
@@ -281,7 +281,7 @@ def __call__(self, x, uttid=None, train=True):
                 # Randomly select the noise source
                 noise = self.state.choice(list(self.utt2noise.values()))
             # Normalize the level
-            noise /= numpy.sqrt((noise ** 2).mean())
+            noise /= numpy.sqrt((noise**2).mean())
 
             # Adjust the noise length
             diff = abs(len(x) - len(noise))

espnet2/asr/frontend/default.py

@@ -109,7 +109,7 @@ def forward(
 
         # 4. STFT -> Power spectrum
         # h: ComplexTensor(B, T, F) -> torch.Tensor(B, T, F)
-        input_power = input_stft.real ** 2 + input_stft.imag ** 2
+        input_power = input_stft.real**2 + input_stft.imag**2
 
         # 5. Feature transform e.g. Stft -> Log-Mel-Fbank
         # input_power: (Batch, [Channel,] Length, Freq)

espnet2/enh/layers/dnn_wpe.py

@@ -97,7 +97,7 @@ def forward(
 
         for i in range(self.iterations):
             # Calculate power: (..., C, T)
-            power = [enh.real ** 2 + enh.imag ** 2 for enh in enhanced]
+            power = [enh.real**2 + enh.imag**2 for enh in enhanced]
             if i == 0 and self.use_dnn_mask:
                 # mask: (B, F, C, T)
                 masks, _ = self.mask_est(data, ilens)

espnet2/enh/layers/mask_estimator.py

@@ -60,7 +60,7 @@ def forward(
 
         # Calculate amplitude: (B, C, T, F) -> (B, C, T, F)
         if is_complex(xs):
-            xs = (xs.real ** 2 + xs.imag ** 2) ** 0.5
+            xs = (xs.real**2 + xs.imag**2) ** 0.5
         # xs: (B, C, T, F) -> xs: (B * C, T, F)
         xs = xs.contiguous().view(-1, xs.size(-2), xs.size(-1))
         # ilens: (B,) -> ilens_: (B * C)

espnet2/enh/layers/tcn.py

@@ -46,7 +46,7 @@ def __init__(
         for r in range(R):
             blocks = []
             for x in range(X):
-                dilation = 2 ** x
+                dilation = 2**x
                 padding = (P - 1) * dilation if causal else (P - 1) * dilation // 2
                 blocks += [
                     TemporalBlock(

espnet2/enh/layers/wpe.py

@@ -65,7 +65,7 @@ def get_power(signal, dim=-2) -> torch.Tensor:
         Power with shape (F, T)
 
     """
-    power = signal.real ** 2 + signal.imag ** 2
+    power = signal.real**2 + signal.imag**2
     power = power.mean(dim=dim)
     return power
 

espnet2/enh/loss/criterions/tf_domain.py

@@ -126,9 +126,9 @@ def forward(self, ref, inf) -> torch.Tensor:
 
         diff = ref - inf
         if is_complex(diff):
-            mseloss = diff.real ** 2 + diff.imag ** 2
+            mseloss = diff.real**2 + diff.imag**2
         else:
-            mseloss = diff ** 2
+            mseloss = diff**2
         if ref.dim() == 3:
             mseloss = mseloss.mean(dim=[1, 2])
         elif ref.dim() == 4:

espnet2/enh/loss/criterions/time_domain.py

@@ -107,15 +107,15 @@ def forward(
         # s_target = <s', s>s / ||s||^2
         pair_wise_dot = torch.sum(s_estimate * s_target, dim=1, keepdim=True)  # [B, 1]
         s_target_energy = (
-            torch.sum(s_target ** 2, dim=1, keepdim=True) + self.eps
+            torch.sum(s_target**2, dim=1, keepdim=True) + self.eps
         )  # [B, 1]
         pair_wise_proj = pair_wise_dot * s_target / s_target_energy  # [B, T]
         # e_noise = s' - s_target
         e_noise = s_estimate - pair_wise_proj  # [B, T]
 
         # SI-SNR = 10 * log_10(||s_target||^2 / ||e_noise||^2)
-        pair_wise_si_snr = torch.sum(pair_wise_proj ** 2, dim=1) / (
-            torch.sum(e_noise ** 2, dim=1) + self.eps
+        pair_wise_si_snr = torch.sum(pair_wise_proj**2, dim=1) / (
+            torch.sum(e_noise**2, dim=1) + self.eps
         )
         pair_wise_si_snr = 10 * torch.log10(pair_wise_si_snr + self.eps)  # [B]
 

espnet2/gan_tts/hifigan/hifigan.py

@@ -93,7 +93,7 @@ def __init__(
                         **nonlinear_activation_params
                     ),
                     torch.nn.ConvTranspose1d(
-                        channels // (2 ** i),
+                        channels // (2**i),
                         channels // (2 ** (i + 1)),
                         upsample_kernel_sizes[i],
                         upsample_scales[i],

espnet2/gan_tts/melgan/melgan.py

@@ -84,7 +84,7 @@ def __init__(
             ]
             layers += [
                 torch.nn.ConvTranspose1d(
-                    channels // (2 ** i),
+                    channels // (2**i),
                     channels // (2 ** (i + 1)),
                     upsample_scale * 2,
                     stride=upsample_scale,
@@ -100,7 +100,7 @@ def __init__(
                     ResidualStack(
                         kernel_size=stack_kernel_size,
                         channels=channels // (2 ** (i + 1)),
-                        dilation=stack_kernel_size ** j,
+                        dilation=stack_kernel_size**j,
                         bias=bias,
                         nonlinear_activation=nonlinear_activation,
                         nonlinear_activation_params=nonlinear_activation_params,

espnet2/gan_tts/parallel_wavegan/parallel_wavegan.py

@@ -196,7 +196,7 @@ def _apply_weight_norm(m: torch.nn.Module):
 
     @staticmethod
     def _get_receptive_field_size(
-        layers, stacks, kernel_size, dilation=lambda x: 2 ** x
+        layers, stacks, kernel_size, dilation=lambda x: 2**x
     ):
         assert layers % stacks == 0
         layers_per_cycle = layers // stacks
@@ -289,7 +289,7 @@ def __init__(
             if i == 0:
                 dilation = 1
             else:
-                dilation = i if dilation_factor == 1 else dilation_factor ** i
+                dilation = i if dilation_factor == 1 else dilation_factor**i
                 conv_in_channels = conv_channels
             padding = (kernel_size - 1) // 2 * dilation
             conv_layer = [

espnet2/gan_tts/vits/duration_predictor.py

@@ -157,7 +157,7 @@ def forward(
                 (F.logsigmoid(z_u) + F.logsigmoid(-z_u)) * x_mask, [1, 2]
             )
             logq = (
-                torch.sum(-0.5 * (math.log(2 * math.pi) + (e_q ** 2)) * x_mask, [1, 2])
+                torch.sum(-0.5 * (math.log(2 * math.pi) + (e_q**2)) * x_mask, [1, 2])
                 - logdet_tot_q
             )
 
@@ -169,7 +169,7 @@ def forward(
                 z, logdet = flow(z, x_mask, g=x, inverse=inverse)
                 logdet_tot = logdet_tot + logdet
             nll = (
-                torch.sum(0.5 * (math.log(2 * math.pi) + (z ** 2)) * x_mask, [1, 2])
+                torch.sum(0.5 * (math.log(2 * math.pi) + (z**2)) * x_mask, [1, 2])
                 - logdet_tot
             )
             return nll + logq  # (B,)

espnet2/gan_tts/vits/flow.py

@@ -155,7 +155,7 @@ def __init__(
 
         self.convs = torch.nn.ModuleList()
         for i in range(layers):
-            dilation = kernel_size ** i
+            dilation = kernel_size**i
             padding = (kernel_size * dilation - dilation) // 2
             self.convs += [
                 torch.nn.Sequential(