Remove SpeedySpeech from .models.json

TTS/.models.json

@@ -47,15 +47,6 @@
                     "license": "MPL",
                     "contact": "egolge@coqui.com"
                 },
-                "speedy-speech-wn": {
-                    "description": "Speedy Speech model with wavenet decoder.",
-                    "github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.1.0/tts_models--en--ljspeech--speedy-speech-wn.zip",
-                    "default_vocoder": "vocoder_models/en/ljspeech/multiband-melgan",
-                    "commit": "77b6145",
-                    "author": "Eren Gölge @erogol",
-                    "license": "MPL",
-                    "contact": "egolge@coqui.com"
-                },
                 "vits": {
                     "description": "VITS is an End2End TTS model trained on LJSpeech dataset with phonemes.",
                     "github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.2.0/tts_models--en--ljspeech--vits.zip",

TTS/tts/layers/__init__.py

@@ -1,15 +1 @@
 from TTS.tts.layers.losses import *
-
-
-def setup_loss(config):
-    if config.model.lower() in ["tacotron", "tacotron2"]:
-        model = TacotronLoss(config)
-    elif config.model.lower() == "glow_tts":
-        model = GlowTTSLoss()
-    elif config.model.lower() == "speedy_speech":
-        model = SpeedySpeechLoss(config)
-    elif config.model.lower() == "align_tts":
-        model = AlignTTSLoss(config)
-    else:
-        raise ValueError(f" [!] loss for model {config.model.lower()} cannot be found.")
-    return model

TTS/tts/utils/data.py

@@ -1,5 +1,4 @@
 import numpy as np
-import torch
 
 
 def _pad_data(x, length):

TTS/tts/utils/helpers.py

@@ -11,11 +11,11 @@ except ModuleNotFoundError:
 
 
 class StandardScaler:
-    """StandardScaler for mean-std normalization with the given mean and std values."""
+    """StandardScaler for mean-scale normalization with the given mean and scale values."""
 
-    def __init__(self, mean: np.ndarray = None, std: np.ndarray = None) -> None:
+    def __init__(self, mean: np.ndarray = None, scale: np.ndarray = None) -> None:
         self.mean_ = mean
-        self.std_ = std
+        self.scale_ = scale
 
     def set_stats(self, mean, scale):
         self.mean_ = mean

tests/tts_tests/test_forward_tts.py

@@ -1,5 +1,3 @@
-import unittest
-
 import torch as T
 
 from TTS.tts.models.forward_tts import ForwardTTS, ForwardTTSArgs
@@ -54,12 +52,12 @@ def model_input_output_test():
     assert (outputs["x_mask"] - x_mask).sum() == 0.0
     assert (outputs["y_mask"] - y_mask).sum() == 0.0
 
-    assert outputs["alignment_soft"] == None
+    assert outputs["alignment_soft"] is None
-    assert outputs["alignment_mas"] == None
+    assert outputs["alignment_mas"] is None
-    assert outputs["alignment_logprob"] == None
+    assert outputs["alignment_logprob"] is None
-    assert outputs["o_alignment_dur"] == None
+    assert outputs["o_alignment_dur"] is None
-    assert outputs["pitch_avg"] == None
+    assert outputs["pitch_avg"] is None
-    assert outputs["pitch_avg_gt"] == None
+    assert outputs["pitch_avg_gt"] is None
 
     # USE PITCH
     model = ForwardTTS(ForwardTTSArgs(num_chars=10, use_pitch=True, use_aligner=False))
@@ -85,10 +83,10 @@ def model_input_output_test():
     assert outputs["pitch_avg"].shape == (2, 1, 21)
     assert outputs["pitch_avg_gt"].shape == (2, 1, 21)
 
-    assert outputs["alignment_soft"] == None
+    assert outputs["alignment_soft"] is None
-    assert outputs["alignment_mas"] == None
+    assert outputs["alignment_mas"] is None
-    assert outputs["alignment_logprob"] == None
+    assert outputs["alignment_logprob"] is None
-    assert outputs["o_alignment_dur"] == None
+    assert outputs["o_alignment_dur"] is None
 
     # USE ALIGNER NETWORK
     model = ForwardTTS(ForwardTTSArgs(num_chars=10, use_pitch=False, use_aligner=True))
@@ -116,8 +114,8 @@ def model_input_output_test():
     assert outputs["alignment_logprob"].shape == (2, 1, durations.sum(1).max(), 21)
     assert outputs["o_alignment_dur"].shape == (2, 21)
 
-    assert outputs["pitch_avg"] == None
+    assert outputs["pitch_avg"] is None
-    assert outputs["pitch_avg_gt"] == None
+    assert outputs["pitch_avg_gt"] is None
 
     # USE ALIGNER NETWORK AND PITCH
     model = ForwardTTS(ForwardTTSArgs(num_chars=10, use_pitch=True, use_aligner=True))

tests/tts_tests/test_helpers.py

@@ -1,6 +1,6 @@
 import torch as T
 
-from TTS.tts.utils.helpers import *
+from TTS.tts.utils.helpers import average_over_durations, generate_path, segment, sequence_mask
 
 
 def average_over_durations_test():  # pylint: disable=no-self-use
@@ -47,7 +47,7 @@ def generate_path_test():
     durations = durations * x_mask.squeeze(1)
     y_length = durations.sum(1)
     y_mask = sequence_mask(y_length).unsqueeze(1).long()
-    attn_mask = (torch.unsqueeze(x_mask, -1) * torch.unsqueeze(y_mask, 2)).squeeze(1).long()
+    attn_mask = (T.unsqueeze(x_mask, -1) * T.unsqueeze(y_mask, 2)).squeeze(1).long()
     print(attn_mask.shape)
     path = generate_path(durations, attn_mask)
     assert path.shape == (10, 21, durations.sum(1).max().item())