refactoring tests after Coqpit

run_bash_tests.sh

@@ -2,13 +2,7 @@ set -e
 TF_CPP_MIN_LOG_LEVEL=3
 
 # runtime bash based tests
+# TODO: move these to python
 ./tests/bash_tests/test_demo_server.sh && \
 ./tests/bash_tests/test_resample.sh && \
-./tests/bash_tests/test_tacotron_train.sh && \
-./tests/bash_tests/test_glow-tts_train.sh && \
-./tests/bash_tests/test_vocoder_gan_train.sh && \
-./tests/bash_tests/test_vocoder_wavernn_train.sh && \
-./tests/bash_tests/test_vocoder_wavegrad_train.sh && \
-./tests/bash_tests/test_speedy_speech_train.sh && \
-./tests/bash_tests/test_aligntts_train.sh && \
 ./tests/bash_tests/test_compute_statistics.sh

tests/inputs/test_config.json

@@ -1,24 +1,24 @@
     {
     "audio":{
-        "audio_processor": "audio",     // to use dictate different audio processors, if available.
-        "num_mels": 80,         // size of the mel spec frame.
-        "fft_size": 1024,       // number of stft frequency levels. Size of the linear spectogram frame.
-        "sample_rate": 22050,   // wav sample-rate. If different than the original data, it is resampled.
-        "frame_length_ms": null,  // stft window length in ms.
-        "frame_shift_ms": null, // stft window hop-lengh in ms.
+        "audio_processor": "audio",
+        "num_mels": 80,
+        "fft_size": 1024,
+        "sample_rate": 22050,
+        "frame_length_ms": null,
+        "frame_shift_ms": null,
         "hop_length": 256,
         "win_length": 1024,
-        "preemphasis": 0.97,    // pre-emphasis to reduce spec noise and make it more structured. If 0.0, no -pre-emphasis.
-        "min_level_db": -100,   // normalization range
-        "ref_level_db": 20,     // reference level db, theoretically 20db is the sound of air.
-        "power": 1.5,           // value to sharpen wav signals after GL algorithm.
-        "griffin_lim_iters": 30,// #griffin-lim iterations. 30-60 is a good range. Larger the value, slower the generation.
-        "signal_norm": true,    // normalize the spec values in range [0, 1]
-        "symmetric_norm": true, // move normalization to range [-1, 1]
-        "clip_norm": true,       // clip normalized values into the range.
-        "max_norm": 4,          // scale normalization to range [-max_norm, max_norm] or [0, max_norm]
-        "mel_fmin": 0,         // minimum freq level for mel-spec. ~50 for male and ~95 for female voices. Tune for dataset!!
-        "mel_fmax": 8000,        // maximum freq level for mel-spec. Tune for dataset!!
+        "preemphasis": 0.97,
+        "min_level_db": -100,
+        "ref_level_db": 20,
+        "power": 1.5,
+        "griffin_lim_iters": 30,
+        "signal_norm": true,
+        "symmetric_norm": true,
+        "clip_norm": true,
+        "max_norm": 4,
+        "mel_fmin": 0,
+        "mel_fmax": 8000,
         "do_trim_silence": false,
         "spec_gain": 20
     },
@@ -53,15 +53,15 @@
     "max_seq_len": 300,
     "log_dir": "tests/outputs/",
 
-    // MULTI-SPEAKER and GST
-    "use_speaker_embedding": false,     // use speaker embedding to enable multi-speaker learning.
-    "use_gst": true,       			    // use global style tokens
-    "gst":	{			                // gst parameter if gst is enabled
-        "gst_style_input": null,        // Condition the style input either on a
-                                        // -> wave file [path to wave] or
-                                        // -> dictionary using the style tokens {'token1': 'value', 'token2': 'value'} example {"0": 0.15, "1": 0.15, "5": -0.15}
-                                        // with the dictionary being len(dict) <= len(gst_num_style_tokens).
-        "gst_use_speaker_embedding": true, // if true pass speaker embedding in attention input GST.
+
+    "use_speaker_embedding": false,
+    "use_gst": true,
+    "gst":	{
+        "gst_style_input": null,
+
+
+
+        "gst_use_speaker_embedding": true,
         "gst_embedding_dim": 512,
         "gst_num_heads": 4,
         "gst_num_style_tokens": 10

tests/inputs/test_speaker_encoder_config.json

@@ -1,5 +1,6 @@
 
 {
+    "model": "speaker_encoder",
     "run_name": "test_speaker_encoder",
     "run_description": "test speaker encoder.",
     "audio":{
@@ -42,8 +43,9 @@
     "checkpoint": true, // If true, it saves checkpoints per "save_step"
     "save_step": 1000, // Number of training steps expected to save traning stats and checkpoints.
     "print_step": 20, // Number of steps to log traning on console.
+    "batch_size": 32,
     "output_path": "", // DATASET-RELATED: output path for all training outputs.
-    "model": {
+    "model_params": {
         "input_dim": 40,
         "proj_dim": 256,
         "lstm_dim": 768,

tests/outputs/dummy_model_config.json

@@ -87,7 +87,6 @@
 
     // MULTI-SPEAKER and GST
     "use_speaker_embedding": false,     // use speaker embedding to enable multi-speaker learning.
-    "use_gst": true,       			    // use global style tokens
     "gst":	{			                // gst parameter if gst is enabled
         "gst_style_input": null,        // Condition the style input either on a
                                         // -> wave file [path to wave] or

tests/test_audio.py

@@ -3,21 +3,21 @@
 
 from tests import get_tests_input_path, get_tests_output_path, get_tests_path
 from TTS.utils.audio import AudioProcessor
-from TTS.utils.io import load_config
+from TTS.config import BaseAudioConfig
 
 TESTS_PATH = get_tests_path()
 OUT_PATH = os.path.join(get_tests_output_path(), "audio_tests")
 WAV_FILE = os.path.join(get_tests_input_path(), "example_1.wav")
 
 os.makedirs(OUT_PATH, exist_ok=True)
-conf = load_config(os.path.join(get_tests_input_path(), "test_config.json"))
+conf = BaseAudioConfig(mel_fmax=8000)
 
 
 # pylint: disable=protected-access
 class TestAudio(unittest.TestCase):
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
-        self.ap = AudioProcessor(**conf.audio)
+        self.ap = AudioProcessor(**conf)
 
     def test_audio_synthesis(self):
         """1. load wav
@@ -163,12 +163,12 @@ def test_normalize(self):
 
     def test_scaler(self):
         scaler_stats_path = os.path.join(get_tests_input_path(), "scale_stats.npy")
-        conf.audio["stats_path"] = scaler_stats_path
-        conf.audio["preemphasis"] = 0.0
-        conf.audio["do_trim_silence"] = True
-        conf.audio["signal_norm"] = True
+        conf.stats_path = scaler_stats_path
+        conf.preemphasis = 0.0
+        conf.do_trim_silence = True
+        conf.signal_norm = True
 
-        ap = AudioProcessor(**conf.audio)
+        ap = AudioProcessor(**conf)
         mel_mean, mel_std, linear_mean, linear_std, _ = ap.load_stats(scaler_stats_path)
         ap.setup_scaler(mel_mean, mel_std, linear_mean, linear_std)
 

tests/test_glow_tts.py

@@ -17,7 +17,7 @@
 use_cuda = torch.cuda.is_available()
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
-c = load_config(os.path.join(get_tests_input_path(), "test_config.json"))
+c = GlowTTSConfig()
 
 ap = AudioProcessor(**c.audio)
 WAV_FILE = os.path.join(get_tests_input_path(), "example_1.wav")

tests/test_loader.py

@@ -10,13 +10,17 @@
 from TTS.tts.datasets import TTSDataset
 from TTS.tts.datasets.preprocess import ljspeech
 from TTS.utils.audio import AudioProcessor
-from TTS.utils.io import load_config
+from TTS.tts.configs import BaseTTSConfig
 
 # pylint: disable=unused-variable
 
 OUTPATH = os.path.join(get_tests_output_path(), "loader_tests/")
 os.makedirs(OUTPATH, exist_ok=True)
-c = load_config(os.path.join(get_tests_input_path(), "test_config.json"))
+
+# create a dummy config for testing data loaders.
+c = BaseTTSConfig(text_cleaner='english_cleaners', num_loader_workers=0, batch_size=2)
+c.r = 5
+c.data_path = "tests/data/ljspeech/"
 ok_ljspeech = os.path.exists(c.data_path)
 
 DATA_EXIST = True
@@ -40,7 +44,7 @@ def _create_dataloader(self, batch_size, r, bgs):
             compute_linear_spec=True,
             ap=self.ap,
             meta_data=items,
-            tp=c.characters if "characters" in c.keys() else None,
+            tp=c.characters,
             batch_group_size=bgs,
             min_seq_len=c.min_seq_len,
             max_seq_len=float("inf"),

tests/test_speaker_encoder_train.py

@@ -2,7 +2,8 @@
 import os
 import shutil
 
-from tests import get_tests_output_path, run_cli
+from tests import get_tests_output_path, run_cli, get_device_id
+
 from TTS.speaker_encoder.speaker_encoder_config import SpeakerEncoderConfig
 from TTS.config.shared_configs import BaseAudioConfig
 
@@ -15,9 +16,9 @@
     num_speakers_in_batch=1,
     num_utters_per_speaker=10,
     num_loader_workers=0,
-    max_train_step=10,
+    max_train_step=2,
     print_step=1,
-    save_step=10,
+    save_step=1,
     print_eval=True,
     audio=BaseAudioConfig(num_mels=40)
 )
@@ -27,7 +28,7 @@
 
 # train the model for one epoch
 command_train = (
-    f"CUDA_VISIBLE_DEVICES='' python TTS/bin/train_encoder.py --config_path {config_path} "
+    f"CUDA_VISIBLE_DEVICES='{get_device_id()}' python TTS/bin/train_encoder.py --config_path {config_path} "
     f"--coqpit.output_path {output_path} "
     "--coqpit.datasets.0.name ljspeech "
     "--coqpit.datasets.0.meta_file_train metadata.csv "
@@ -41,6 +42,6 @@
 continue_path = max(glob.glob(os.path.join(output_path, "*/")), key=os.path.getmtime)
 
 # restore the model and continue training for one more epoch
-command_train = f"CUDA_VISIBLE_DEVICES='' python TTS/bin/train_encoder.py --continue_path {continue_path} "
+command_train = f"CUDA_VISIBLE_DEVICES='{get_device_id()}' python TTS/bin/train_encoder.py --continue_path {continue_path} "
 run_cli(command_train)
 shutil.rmtree(continue_path)

tests/test_speaker_manager.py

@@ -26,11 +26,11 @@ class SpeakerManagerTest(unittest.TestCase):
     def test_speaker_embedding():
         # load config
         config = load_config(encoder_config_path)
-        config["audio"]["resample"] = True
+        config.audio.resample = True
 
         # create a dummy speaker encoder
-        model = SpeakerEncoder(**config.model)
-        save_checkpoint(model, None, None, get_tests_input_path(), 0, 0)
+        model = SpeakerEncoder(**config.model_params)
+        save_checkpoint(model, None, None, get_tests_input_path(), 0)
 
         # load audio processor and speaker encoder
         ap = AudioProcessor(**config.audio)

tests/test_synthesize.py

@@ -1,6 +1,7 @@
 import os
 
-from tests import get_tests_output_path, run_cli
+from tests import get_tests_output_path, run_cli, get_device_id
+
 
 
 def test_synthesize():

tests/test_synthesizer.py

@@ -15,8 +15,8 @@ def _create_random_model(self):
         # pylint: disable=global-statement
         global symbols, phonemes
         config = load_config(os.path.join(get_tests_output_path(), "dummy_model_config.json"))
-        if "characters" in config.keys():
-            symbols, phonemes = make_symbols(**config.characters)
+        if config.has('characters') and config.characters:
+            symbols, phonemes = make_symbols(**config.characters.to_dict())
 
         num_chars = len(phonemes) if config.use_phonemes else len(symbols)
         model = setup_model(num_chars, 0, config)
@@ -25,11 +25,10 @@ def _create_random_model(self):
 
     def test_in_out(self):
         self._create_random_model()
-        config = load_config(os.path.join(get_tests_input_path(), "server_config.json"))
         tts_root_path = get_tests_output_path()
-        config["tts_checkpoint"] = os.path.join(tts_root_path, config["tts_checkpoint"])
-        config["tts_config"] = os.path.join(tts_root_path, config["tts_config"])
-        synthesizer = Synthesizer(config["tts_checkpoint"], config["tts_config"], None, None)
+        tts_checkpoint = os.path.join(tts_root_path, 'checkpoint_10.pth.tar')
+        tts_config = os.path.join(tts_root_path, 'dummy_model_config.json')
+        synthesizer = Synthesizer(tts_checkpoint, tts_config, None, None)
         synthesizer.tts("Better this test works!!")
 
     def test_split_into_sentences(self):

tests/test_tacotron2_model.py

@@ -17,7 +17,7 @@
 use_cuda = torch.cuda.is_available()
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
-c = load_config(os.path.join(get_tests_input_path(), "test_config.json"))
+c = Tacotron2Config()
 
 ap = AudioProcessor(**c.audio)
 WAV_FILE = os.path.join(get_tests_input_path(), "example_1.wav")
@@ -152,10 +152,8 @@ def test_train_step(self):
             num_chars=24,
             r=c.r,
             num_speakers=5,
-            gst=True,
-            gst_embedding_dim=c.gst["gst_embedding_dim"],
-            gst_num_heads=c.gst["gst_num_heads"],
-            gst_style_tokens=c.gst["gst_style_tokens"],
+            use_gst=True,
+            gst=c.gst
         ).to(device)
         model.train()
         model_ref = copy.deepcopy(model)
@@ -216,10 +214,8 @@ def test_train_step(self):
             num_chars=24,
             r=c.r,
             num_speakers=5,
-            gst=True,
-            gst_embedding_dim=c.gst["gst_embedding_dim"],
-            gst_num_heads=c.gst["gst_num_heads"],
-            gst_style_tokens=c.gst["gst_style_tokens"],
+            use_gst=True,
+            gst =c.gst
         ).to(device)
         model.train()
         model_ref = copy.deepcopy(model)
@@ -280,11 +276,8 @@ def test_train_step():
             r=c.r,
             num_speakers=5,
             speaker_embedding_dim=55,
-            gst=True,
-            gst_embedding_dim=c.gst["gst_embedding_dim"],
-            gst_num_heads=c.gst["gst_num_heads"],
-            gst_style_tokens=c.gst["gst_style_tokens"],
-            gst_use_speaker_embedding=c.gst["gst_use_speaker_embedding"],
+            use_gst=True,
+            gst=c.gst
         ).to(device)
         model.train()
         model_ref = copy.deepcopy(model)

tests/test_tacotron2_tf_model.py

@@ -19,7 +19,7 @@
 use_cuda = torch.cuda.is_available()
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
-c = load_config(os.path.join(get_tests_input_path(), "test_config.json"))
+c = Tacotron2Config()
 
 
 class TacotronTFTrainTest(unittest.TestCase):

tests/test_tacotron_model.py

@@ -18,7 +18,7 @@
 use_cuda = torch.cuda.is_available()
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
-c = load_config(os.path.join(get_tests_input_path(), "test_config.json"))
+c = TacotronConfig()
 
 ap = AudioProcessor(**c.audio)
 WAV_FILE = os.path.join(get_tests_input_path(), "example_1.wav")
@@ -175,10 +175,8 @@ def test_train_step():
         model = Tacotron(
             num_chars=32,
             num_speakers=5,
-            gst=True,
-            gst_embedding_dim=c.gst["gst_embedding_dim"],
-            gst_num_heads=c.gst["gst_num_heads"],
-            gst_style_tokens=c.gst["gst_style_tokens"],
+            use_gst=True,
+            gst=c.gst,
             postnet_output_dim=c.audio["fft_size"],
             decoder_output_dim=c.audio["num_mels"],
             r=c.r,
@@ -240,10 +238,8 @@ def test_train_step():
         model = Tacotron(
             num_chars=32,
             num_speakers=5,
-            gst=True,
-            gst_embedding_dim=c.gst["gst_embedding_dim"],
-            gst_num_heads=c.gst["gst_num_heads"],
-            gst_style_tokens=c.gst["gst_style_tokens"],
+            use_gst=True,
+            gst=c.gst,
             postnet_output_dim=c.audio["fft_size"],
             decoder_output_dim=c.audio["num_mels"],
             r=c.r,
@@ -306,11 +302,8 @@ def test_train_step():
             num_speakers=5,
             postnet_output_dim=c.audio["fft_size"],
             decoder_output_dim=c.audio["num_mels"],
-            gst=True,
-            gst_embedding_dim=c.gst["gst_embedding_dim"],
-            gst_num_heads=c.gst["gst_num_heads"],
-            gst_style_tokens=c.gst["gst_style_tokens"],
-            gst_use_speaker_embedding=c.gst["gst_use_speaker_embedding"],
+            use_gst=True,
+            gst=c.gst,
             r=c.r,
             memory_size=c.memory_size,
             speaker_embedding_dim=55,