Merge branch 'dev' of https://github.com/mozilla/TTS into dev

TTS/bin/convert_melgan_torch_to_tf.py

@@ -1,10 +1,10 @@
 import argparse
+from difflib import SequenceMatcher
 import os
 
 import numpy as np
 import tensorflow as tf
 import torch
-from fuzzywuzzy import fuzz
 
 from TTS.utils.io import load_config
 from TTS.vocoder.tf.utils.convert_torch_to_tf_utils import (
@@ -67,7 +67,7 @@ for tf_name in tf_var_names:
         continue
     tf_name_edited = convert_tf_name(tf_name)
     ratios = [
-        fuzz.ratio(torch_name, tf_name_edited)
+        SequenceMatcher(None, torch_name, tf_name_edited).ratio()
         for torch_name in torch_var_names
     ]
     max_idx = np.argmax(ratios)

TTS/bin/convert_tacotron2_torch_to_tf.py

@@ -1,6 +1,7 @@
 # %%
 # %%
 import argparse
+from difflib import SequenceMatcher
 import os
 import sys
 # %%
@@ -10,7 +11,6 @@ from pprint import pprint
 import numpy as np
 import tensorflow as tf
 import torch
-from fuzzywuzzy import fuzz
 from TTS.tts.tf.models.tacotron2 import Tacotron2
 from TTS.tts.tf.utils.convert_torch_to_tf_utils import (
     compare_torch_tf, convert_tf_name, transfer_weights_torch_to_tf)
@@ -106,7 +106,7 @@ for tf_name in tf_var_names:
         continue
     tf_name_edited = convert_tf_name(tf_name)
     ratios = [
-        fuzz.ratio(torch_name, tf_name_edited)
+        SequenceMatcher(None, torch_name, tf_name_edited).ratio()
         for torch_name in torch_var_names
     ]
     max_idx = np.argmax(ratios)

TTS/bin/train_encoder.py

@@ -42,8 +42,12 @@ def setup_loader(ap, is_val=False, verbose=False):
         dataset = MyDataset(ap,
                             meta_data_eval if is_val else meta_data_train,
                             voice_len=1.6,
-                            num_utter_per_speaker=10,
+                            num_utter_per_speaker=c.num_utters_per_speaker,
+                            num_speakers_in_batch=c.num_speakers_in_batch,
                             skip_speakers=False,
+                            storage_size=c.storage["storage_size"],
+                            sample_from_storage_p=c.storage["sample_from_storage_p"],
+                            additive_noise=c.storage["additive_noise"],
                             verbose=verbose)
         # sampler = DistributedSampler(dataset) if num_gpus > 1 else None
         loader = DataLoader(dataset,
@@ -60,6 +64,7 @@ def train(model, criterion, optimizer, scheduler, ap, global_step):
     epoch_time = 0
     best_loss = float('inf')
     avg_loss = 0
+    avg_loader_time = 0
     end_time = time.time()
     for _, data in enumerate(data_loader):
         start_time = time.time()
@@ -93,8 +98,11 @@ def train(model, criterion, optimizer, scheduler, ap, global_step):
         step_time = time.time() - start_time
         epoch_time += step_time
 
-        avg_loss = 0.01 * loss.item(
+        # Averaged Loss and Averaged Loader Time
-        ) + 0.99 * avg_loss if avg_loss != 0 else loss.item()
+        avg_loss = 0.01 * loss.item() \
+                   + 0.99 * avg_loss if avg_loss != 0 else loss.item()
+        avg_loader_time = 1/c.num_loader_workers * loader_time + \
+                          (c.num_loader_workers-1) / c.num_loader_workers * avg_loader_time if avg_loader_time != 0 else loader_time
         current_lr = optimizer.param_groups[0]['lr']
 
         if global_step % c.steps_plot_stats == 0:
@@ -103,7 +111,8 @@ def train(model, criterion, optimizer, scheduler, ap, global_step):
                 "loss": avg_loss,
                 "lr": current_lr,
                 "grad_norm": grad_norm,
-                "step_time": step_time
+                "step_time": step_time,
+                "avg_loader_time": avg_loader_time
             }
             tb_logger.tb_train_epoch_stats(global_step, train_stats)
             figures = {
@@ -116,9 +125,9 @@ def train(model, criterion, optimizer, scheduler, ap, global_step):
         if global_step % c.print_step == 0:
             print(
                 "   | > Step:{}  Loss:{:.5f}  AvgLoss:{:.5f}  GradNorm:{:.5f}  "
-                "StepTime:{:.2f}  LoaderTime:{:.2f}  LR:{:.6f}".format(
+                "StepTime:{:.2f}  LoaderTime:{:.2f}  AvGLoaderTime:{:.2f}  LR:{:.6f}".format(
                     global_step, loss.item(), avg_loss, grad_norm, step_time,
-                    loader_time, current_lr),
+                    loader_time, avg_loader_time, current_lr),
                 flush=True)
 
         # save best model

TTS/speaker_encoder/dataset.py

@@ -1,11 +1,15 @@
+import numpy
 import numpy as np
+import queue
 import torch
 import random
 from torch.utils.data import Dataset
+from tqdm import tqdm
 
 
 class MyDataset(Dataset):
     def __init__(self, ap, meta_data, voice_len=1.6, num_speakers_in_batch=64,
+                 storage_size=1, sample_from_storage_p=0.5, additive_noise=0,
                  num_utter_per_speaker=10, skip_speakers=False, verbose=False):
         """
         Args:
@@ -24,8 +28,15 @@ class MyDataset(Dataset):
         self.ap = ap
         self.verbose = verbose
         self.__parse_items()
+        self.storage = queue.Queue(maxsize=storage_size*num_speakers_in_batch)
+        self.sample_from_storage_p = float(sample_from_storage_p)
+        self.additive_noise = float(additive_noise)
         if self.verbose:
             print("\n > DataLoader initialization")
+            print(f" | > Speakers per Batch: {num_speakers_in_batch}")
+            print(f" | > Storage Size: {self.storage.maxsize} speakers, each with {num_utter_per_speaker} utters")
+            print(f" | > Sample_from_storage_p : {self.sample_from_storage_p}")
+            print(f" | > Noise added : {self.additive_noise}")
             print(f" | > Number of instances : {len(self.items)}")
             print(f" | > Sequence length: {self.seq_len}")
             print(f" | > Num speakers: {len(self.speakers)}")
@@ -51,21 +62,37 @@ class MyDataset(Dataset):
         return sample
 
     def __parse_items(self):
-        """
-        Find unique speaker ids and create a dict mapping utterances from speaker id
-        """
-        speakers = list({item[-1] for item in self.items})
         self.speaker_to_utters = {}
-        self.speakers = []
+        for i in self.items:
-        for speaker in speakers:
+            path_ = i[1]
-            speaker_utters = [item[1] for item in self.items if item[2] == speaker]
+            speaker_ = i[2]
-            if len(speaker_utters) < self.num_utter_per_speaker and self.skip_speakers:
+            if speaker_ in self.speaker_to_utters.keys():
-                print(
+                self.speaker_to_utters[speaker_].append(path_)
-                    f" [!] Skipped speaker {speaker}. Not enough utterances {self.num_utter_per_speaker} vs {len(speaker_utters)}."
-                )
             else:
-                self.speakers.append(speaker)
+                self.speaker_to_utters[speaker_] = [path_, ]
-                self.speaker_to_utters[speaker] = speaker_utters
+
+        if self.skip_speakers:
+            self.speaker_to_utters = {k: v for (k, v) in self.speaker_to_utters.items() if
+                                      len(v) >= self.num_utter_per_speaker}
+
+        self.speakers = [k for (k, v) in self.speaker_to_utters.items()]
+
+    # def __parse_items(self):
+    #     """
+    #     Find unique speaker ids and create a dict mapping utterances from speaker id
+    #     """
+    #     speakers = list({item[-1] for item in self.items})
+    #     self.speaker_to_utters = {}
+    #     self.speakers = []
+    #     for speaker in speakers:
+    #         speaker_utters = [item[1] for item in self.items if item[2] == speaker]
+    #         if len(speaker_utters) < self.num_utter_per_speaker and self.skip_speakers:
+    #             print(
+    #                 f" [!] Skipped speaker {speaker}. Not enough utterances {self.num_utter_per_speaker} vs {len(speaker_utters)}."
+    #             )
+    #         else:
+    #             self.speakers.append(speaker)
+    #             self.speaker_to_utters[speaker] = speaker_utters
 
     def __len__(self):
         return int(1e10)
@@ -86,7 +113,7 @@ class MyDataset(Dataset):
         """
         Sample all M utterances for the given speaker.
         """
-        feats = []
+        wavs = []
         labels = []
         for _ in range(self.num_utter_per_speaker):
             # TODO:dummy but works
@@ -102,11 +129,9 @@ class MyDataset(Dataset):
                     break
                 self.speaker_to_utters[speaker].remove(utter)
 
-            offset = random.randint(0, wav.shape[0] - self.seq_len)
+            wavs.append(wav)
-            mel = self.ap.melspectrogram(wav[offset : offset + self.seq_len])
-            feats.append(torch.FloatTensor(mel))
             labels.append(speaker)
-        return feats, labels
+        return wavs, labels
 
     def __getitem__(self, idx):
         speaker, _ = self.__sample_speaker()
@@ -116,7 +141,28 @@ class MyDataset(Dataset):
         labels = []
         feats = []
         for speaker in batch:
-            feats_, labels_ = self.__sample_speaker_utterances(speaker)
+            if random.random() < self.sample_from_storage_p and self.storage.full():
+                # sample from storage (if full), ignoring the speaker
+                wavs_, labels_ = random.choice(self.storage.queue)
+            else:
+                # don't sample from storage, but from HDD
+                wavs_, labels_ = self.__sample_speaker_utterances(speaker)
+                # if storage is full, remove an item
+                if self.storage.full():
+                    _ = self.storage.get_nowait()
+                # put the newly loaded item into storage
+                self.storage.put_nowait((wavs_, labels_))
+
+            # add random gaussian noise
+            if self.additive_noise > 0:
+                noises_ = [numpy.random.normal(0, self.additive_noise, size=len(w)) for w in wavs_]
+                wavs_ = [wavs_[i] + noises_[i] for i in range(len(wavs_))]
+
+            # get a random subset of each of the wavs and convert to MFCC.
+            offsets_ = [random.randint(0, wav.shape[0] - self.seq_len) for wav in wavs_]
+            mels_ = [self.ap.melspectrogram(wavs_[i][offsets_[i]: offsets_[i] + self.seq_len]) for i in range(len(wavs_))]
+            feats_ = [torch.FloatTensor(mel) for mel in mels_]
+
             labels.append(labels_)
             feats.extend(feats_)
         feats = torch.stack(feats)

TTS/tts/datasets/preprocess.py

@@ -2,6 +2,10 @@ import os
 from glob import glob
 import re
 import sys
+from pathlib import Path
+
+from tqdm import tqdm
+
 from TTS.tts.utils.generic_utils import split_dataset
 
 
@@ -14,8 +18,8 @@ def load_meta_data(datasets):
         meta_file_train = dataset['meta_file_train']
         meta_file_val = dataset['meta_file_val']
         preprocessor = get_preprocessor_by_name(name)
-
         meta_data_train = preprocessor(root_path, meta_file_train)
+        print(f" | > Found {len(meta_data_train)} files in {Path(root_path).resolve()}")
         if meta_file_val is None:
             meta_data_eval, meta_data_train = split_dataset(meta_data_train)
         else:
@@ -167,8 +171,8 @@ def common_voice(root_path, meta_file):
             cols = line.split("\t")
             text = cols[2]
             speaker_name = cols[0]
-            wav_file = os.path.join(root_path, "clips", cols[1] + ".wav")
+            wav_file = os.path.join(root_path, "clips", cols[1].replace(".mp3", ".wav"))
-            items.append([text, wav_file, speaker_name])
+            items.append([text, wav_file, 'MCV_' + speaker_name])
     return items
 
 
@@ -187,7 +191,7 @@ def libri_tts(root_path, meta_files=None):
                 cols = line.split('\t')
                 wav_file = os.path.join(_root_path, cols[0] + '.wav')
                 text = cols[1]
-                items.append([text, wav_file, speaker_name])
+                items.append([text, wav_file, 'LTTS_' + speaker_name])
     for item in items:
         assert os.path.exists(
             item[1]), f" [!] wav files don't exist - {item[1]}"
@@ -235,8 +239,7 @@ def vctk(root_path, meta_files=None, wavs_path='wav48'):
     """homepages.inf.ed.ac.uk/jyamagis/release/VCTK-Corpus.tar.gz"""
     test_speakers = meta_files
     items = []
-    meta_files = glob(f"{os.path.join(root_path,'txt')}/**/*.txt",
+    meta_files = glob(f"{os.path.join(root_path,'txt')}/**/*.txt", recursive=True)
-                      recursive=True)
     for meta_file in meta_files:
         _, speaker_id, txt_file = os.path.relpath(meta_file,
                                                   root_path).split(os.sep)
@@ -249,6 +252,70 @@ def vctk(root_path, meta_files=None, wavs_path='wav48'):
             text = file_text.readlines()[0]
         wav_file = os.path.join(root_path, wavs_path, speaker_id,
                                 file_id + '.wav')
-        items.append([text, wav_file, speaker_id])
+        items.append([text, wav_file, 'VCTK_' + speaker_id])
 
     return items
+
+
+def vctk_slim(root_path, meta_files=None, wavs_path='wav48'):
+    """homepages.inf.ed.ac.uk/jyamagis/release/VCTK-Corpus.tar.gz"""
+    items = []
+    txt_files = glob(f"{os.path.join(root_path,'txt')}/**/*.txt", recursive=True)
+    for text_file in txt_files:
+        _, speaker_id, txt_file = os.path.relpath(text_file,
+                                                  root_path).split(os.sep)
+        file_id = txt_file.split('.')[0]
+        if isinstance(meta_files, list):  # if is list ignore this speakers ids
+            if speaker_id in meta_files:
+                continue
+        wav_file = os.path.join(root_path, wavs_path, speaker_id,
+                                file_id + '.wav')
+        items.append([None, wav_file, 'VCTK_' + speaker_id])
+
+    return items
+
+# ======================================== VOX CELEB ===========================================
+def voxceleb2(root_path, meta_file=None):
+    """
+    :param meta_file   Used only for consistency with load_meta_data api
+    """
+    return _voxcel_x(root_path, meta_file, voxcel_idx="2")
+
+
+def voxceleb1(root_path, meta_file=None):
+    """
+    :param meta_file   Used only for consistency with load_meta_data api
+    """
+    return _voxcel_x(root_path, meta_file, voxcel_idx="1")
+
+
+def _voxcel_x(root_path, meta_file, voxcel_idx):
+    assert voxcel_idx in ["1", "2"]
+    expected_count = 148_000 if voxcel_idx == "1" else 1_000_000
+    voxceleb_path = Path(root_path)
+    cache_to = voxceleb_path / f"metafile_voxceleb{voxcel_idx}.csv"
+    cache_to.parent.mkdir(exist_ok=True)
+
+    # if not exists meta file, crawl recursively for 'wav' files
+    if meta_file is not None:
+        with open(str(meta_file), 'r') as f:
+            return [x.strip().split('|') for x in f.readlines()]
+
+    elif not cache_to.exists():
+        cnt = 0
+        meta_data = ""
+        wav_files = voxceleb_path.rglob("**/*.wav")
+        for path in tqdm(wav_files, desc=f"Building VoxCeleb {voxcel_idx} Meta file ... this needs to be done only once.",
+                         total=expected_count):
+            speaker_id = str(Path(path).parent.parent.stem)
+            assert speaker_id.startswith('id')
+            text = None  # VoxCel does not provide transciptions, and they are not needed for training the SE
+            meta_data += f"{text}|{path}|voxcel{voxcel_idx}_{speaker_id}\n"
+            cnt += 1
+        with open(str(cache_to), 'w') as f:
+            f.write(meta_data)
+        if cnt < expected_count:
+            raise ValueError(f"Found too few instances for Voxceleb. Should be around {expected_count}, is: {cnt}")
+
+    with open(str(cache_to), 'r') as f:
+        return [x.strip().split('|') for x in f.readlines()]

TTS/tts/utils/generic_utils.py

@@ -7,11 +7,9 @@ from TTS.utils.generic_utils import check_argument
 
 
 def split_dataset(items):
-    is_multi_speaker = False
     speakers = [item[-1] for item in items]
     is_multi_speaker = len(set(speakers)) > 1
-    eval_split_size = 500 if len(items) * 0.01 > 500 else int(
+    eval_split_size = min(500, int(len(items) * 0.01))
-        len(items) * 0.01)
     assert eval_split_size > 0, " [!] You do not have enough samples to train. You need at least 100 samples."
     np.random.seed(0)
     np.random.shuffle(items)
@@ -142,6 +140,11 @@ def check_config(c):
     check_argument('do_trim_silence', c['audio'], restricted=True, val_type=bool)
     check_argument('trim_db', c['audio'], restricted=True, val_type=int)
 
+    # storage parameters
+    check_argument('sample_from_storage_p', c['storage'], restricted=True, val_type=float, min_val=0.0, max_val=1.0)
+    check_argument('storage_size', c['storage'], restricted=True, val_type=int, min_val=1, max_val=100)
+    check_argument('additive_noise', c['storage'], restricted=True, val_type=float, min_val=0.0, max_val=1.0)
+
     # training parameters
     check_argument('batch_size', c, restricted=True, val_type=int, min_val=1)
     check_argument('eval_batch_size', c, restricted=True, val_type=int, min_val=1)

TTS/tts/utils/io.py

@@ -50,7 +50,7 @@ def save_best_model(target_loss, best_loss, model, optimizer, current_step, epoc
     if target_loss < best_loss:
         file_name = 'best_model.pth.tar'
         checkpoint_path = os.path.join(output_folder, file_name)
-        print(" > BEST MODEL : {}".format(checkpoint_path))
+        print(" >> BEST MODEL : {}".format(checkpoint_path))
         save_model(model, optimizer, current_step, epoch, r, checkpoint_path, model_loss=target_loss, **kwargs)
         best_loss = target_loss
     return best_loss

requirements.txt

@@ -20,5 +20,4 @@ soundfile
 nose==1.3.7
 cardboardlint==1.3.0
 pylint==2.5.3
-fuzzywuzzy
 gdown

tests/inputs/common_voice.tsv

@@ -1,10 +1,6 @@
-client_id	path	sentence	up_votes	down_votes	age	gender	accent
+client_id	path	sentence	up_votes	down_votes	age	gender	accent	locale	segment
-aa7af576605fee2c78c26b85497c64cb9c9fd97228071f8666d9f49f15bce01899bbb930fa60b76d212091d779d83b92e0b54c73cbb21d2c7e1eedc817e41cb3	21fce545b24d9a5af0403b949e95e8dd3c10c4ff3e371f14e4d5b4ebf588670b7c9e618285fc872d94a89ed7f0217d9019fe5de33f1577b49dcd518eacf63c4b	Man sollte den Länderfinanzausgleich durch einen Bundesliga-Soli ersetzen.	2	0	fourties	male	germany
+95324d489b122a800b840e0b0d068f7363a1a6c2cd2e7365672cc7033e38deaa794bd59edcf8196aa35c9791652b9085ac3839a98bb50ebab4a1e8538a94846b	common_voice_en_20005954.mp3	The applicants are invited for coffee and visa is given immediately.	3	0				en	
-aa7af576605fee2c78c26b85497c64cb9c9fd97228071f8666d9f49f15bce01899bbb930fa60b76d212091d779d83b92e0b54c73cbb21d2c7e1eedc817e41cb3	42758baa4e91ef6b82b78b11a04bc5117a035a8d3bc42c33c0bb3084909af17043a194cfd8cd9839f0d6ef1ea5413acda5de5d1936abcc8ca073e2da7f9488ea	Folgende Lektüre kann ich Ihnen zum Thema Kognitionspsychologie empfehlen.	2	0	fourties	male	germany
+95324d489b122a800b840e0b0d068f7363a1a6c2cd2e7365672cc7033e38deaa794bd59edcf8196aa35c9791652b9085ac3839a98bb50ebab4a1e8538a94846b	common_voice_en_20005955.mp3	Developmental robotics is related to, but differs from, evolutionary robotics.	2	0				en	
-aa7af576605fee2c78c26b85497c64cb9c9fd97228071f8666d9f49f15bce01899bbb930fa60b76d212091d779d83b92e0b54c73cbb21d2c7e1eedc817e41cb3	478f172c2dbda6675247e9674ade79a5b49efeefb7c9e99040dcc69a847a01d69398cf180570859b0cdb6fc887717e04cd8b149c723d48d00b5d18f41314667c	Touristen winkten den Leuten am Ufer zu.	2	0	fourties	male	germany
+95324d489b122a800b840e0b0d068f7363a1a6c2cd2e7365672cc7033e38deaa794bd59edcf8196aa35c9791652b9085ac3839a98bb50ebab4a1e8538a94846b	common_voice_en_20005956.mp3	The musical was originally directed and choreographed by Alan Lund.	2	0				en	
-aa7af576605fee2c78c26b85497c64cb9c9fd97228071f8666d9f49f15bce01899bbb930fa60b76d212091d779d83b92e0b54c73cbb21d2c7e1eedc817e41cb3	4854368d6d21cb44103e432b5332f31e8d14030582a40850501bcf9377d699314a5ff27a8206fa89254ddde7f3f1c65d33836f3dfcfa16bcabec08537f2b5f08	Valentin hat das Handtuch geworfen.	2	0	fourties	male	germany
+954a4181ae9fba89d1b1570f2ae148b3ee18ee2311de978e698f598db859f830d93d35574596d713518e8c96cdae01fce7a08c60c2e0a22bcf01e020924440a6	common_voice_en_19737073.mp3	He graduated from Columbia High School, in Brown County, South Dakota.	2	0				en	
-aa7af576605fee2c78c26b85497c64cb9c9fd97228071f8666d9f49f15bce01899bbb930fa60b76d212091d779d83b92e0b54c73cbb21d2c7e1eedc817e41cb3	a841a9f3e032495dd47560e65fba99eeacb3618c07de8b1351c20188e5b71e33cc52f73315f721a3a24b65763c65bb52fbf3ae052eb5774e834dcb57f296db5c	Ohne Gehörschutz bei der Arbeit wäre Klaus wohl nach zwei Wochen taub.	2	0	fourties	male	germany
+954a4181ae9fba89d1b1570f2ae148b3ee18ee2311de978e698f598db859f830d93d35574596d713518e8c96cdae01fce7a08c60c2e0a22bcf01e020924440a6	common_voice_en_19737074.mp3	Competition for limited resources has also resulted in some local conflicts.	2	0				en	
-aa7af576605fee2c78c26b85497c64cb9c9fd97228071f8666d9f49f15bce01899bbb930fa60b76d212091d779d83b92e0b54c73cbb21d2c7e1eedc817e41cb3	03ab970a5bf5410bc3260b073cce1c7f49c688ace83dc8836b1c0f79a09fea45a27725c769f4a9d2e6181defd016d22642789d7ac51da252b42958a9192bd4c7	Gerrit erinnerte sich daran, dass er einst einen Eid geschworen hatte.	2	0	fourties	male	germany
-aa7af576605fee2c78c26b85497c64cb9c9fd97228071f8666d9f49f15bce01899bbb930fa60b76d212091d779d83b92e0b54c73cbb21d2c7e1eedc817e41cb3	c4a94df443ad5f2c7241413ef7145d5f0de41ae929759073917fe96166da3c7d3a612c920ed7b0f3d5950a38d6205e9dba24af5bfb27e390a220d004e6e26744	Auf das, was jetzt kommt, habe ich nämlich absolut keinen Bock.	2	0	fourties	male	germany
-aa7af576605fee2c78c26b85497c64cb9c9fd97228071f8666d9f49f15bce01899bbb930fa60b76d212091d779d83b92e0b54c73cbb21d2c7e1eedc817e41cb3	104695983b1112229b4a48696405d044dad9ddef713aa6eb1a6240cc16b7b7a2a96354ae9da99783850dde08a982091e48d3037288a3a58269cac9fe70a6bd7a	Von Salzburg ist es doch nicht weit bis zum Chiemsee.	2	0	fourties	male	germany
-d5b5da343bb0f65e3580bc2e1902a4f5d004241488d751503f2020bc1c93f89715e355e35f6e25def2b90cb3eea99fda403eb92ae3afbb84d039a54a4ed2d875	ad2f69e053b0e20e01c82b9821fe5787f1cc8e4b0b97f0e4cab1e9a652c577169c8244fb222281a60ee3081854014113e04c4ca43643100b7c01dab0fac11974	Warum werden da keine strafrechtlichen Konsequenzen gezogen?	2	0	thirties	male	germany

tests/test_preprocessors.py

@@ -11,18 +11,8 @@ class TestPreprocessors(unittest.TestCase):
         root_path = get_tests_input_path()
         meta_file = "common_voice.tsv"
         items = common_voice(root_path, meta_file)
-        assert items[0][0] == "Man sollte den Länderfinanzausgleich durch " \
+        assert items[0][0] == 'The applicants are invited for coffee and visa is given immediately.'
-                              "einen Bundesliga-Soli ersetzen."
+        assert items[0][1] == os.path.join(get_tests_input_path(), "clips", "common_voice_en_20005954.wav")
-        assert items[0][1] == os.path.join(get_tests_input_path(), "clips",
-                                           "21fce545b24d9a5af0403b949e95e8dd3"
-                                           "c10c4ff3e371f14e4d5b4ebf588670b7c"
-                                           "9e618285fc872d94a89ed7f0217d9019f"
-                                           "e5de33f1577b49dcd518eacf63c4b.wav")
 
-        assert items[-1][0] == "Warum werden da keine strafrechtlichen " \
+        assert items[-1][0] == "Competition for limited resources has also resulted in some local conflicts."
-                               "Konsequenzen gezogen?"
+        assert items[-1][1] == os.path.join(get_tests_input_path(), "clips", "common_voice_en_19737074.wav")
-        assert items[-1][1] == os.path.join(get_tests_input_path(), "clips",
-                                            "ad2f69e053b0e20e01c82b9821fe5787f1"
-                                            "cc8e4b0b97f0e4cab1e9a652c577169c82"
-                                            "44fb222281a60ee3081854014113e04c4c"
-                                            "a43643100b7c01dab0fac11974.wav")