Merge pull request #7 from babysor/newdataset

Supporting new dataset SLR68 ! try python synthesizer_preprocess_audi…

README-CN.md

@@ -7,7 +7,7 @@
 ### [English](README.md)  | 中文
 
 ## 特性
-🌍 **中文** 支持普通话并使用数据集进行测试：adatatang_200zh
+🌍 **中文** 支持普通话并使用多种中文数据集进行测试：adatatang_200zh, SLR68
 
 🤩 **PyTorch** 适用于 pytorch，已在 1.9.0 版本（最新于 2021 年 8 月）中测试，GPU Tesla T4 和 GTX 2060
 
@@ -33,6 +33,7 @@ https://github.com/CorentinJ/Real-Time-Voice-Cloning/wiki/Pretrained-models
 * 下载 adatatang_200zh 数据集并解压：确保您可以访问 *train* 文件夹中的所有 .wav
 * 使用音频和梅尔频谱图进行预处理：
 `python synthesizer_preprocess_audio.py <datasets_root>`
+可以传入参数 --dataset `{dataset}` 支持 adatatang_200zh, SLR68
 
 * 预处理嵌入：
 `python synthesizer_preprocess_embeds.py <datasets_root>/SV2TTS/synthesizer`
@@ -48,8 +49,8 @@ https://github.com/CorentinJ/Real-Time-Voice-Cloning/wiki/Pretrained-models
 然后您可以尝试使用工具箱：
 `python demo_toolbox.py -d <datasets_root>`
 
-＃＃ TODO
-- 添加演示视频
-- 添加对更多数据集的支持
-- 上传预训练模型
-- 🙏 欢迎补充
+## TODO
+- [X]  添加演示视频
+- [X] 添加对更多数据集的支持
+- [ ] 上传预训练模型
+- [ ] 🙏 欢迎补充

README.md

@@ -6,7 +6,7 @@
 > English | [中文](README-CN.md) 
 
 ## Features
-🌍 **Chinese** supported mandarin and tested with dataset: aidatatang_200zh
+🌍 **Chinese** supported mandarin and tested with multiple datasets: aidatatang_200zh, SLR68
 
 🤩 **PyTorch** worked for pytorch, tested in version of 1.9.0(latest in August 2021), with GPU Tesla T4 and GTX 2060
 
@@ -35,7 +35,7 @@ https://github.com/CorentinJ/Real-Time-Voice-Cloning/wiki/Pretrained-models
 * Download aidatatang_200zh dataset and unzip: make sure you can access all .wav in *train* folder
 * Preprocess with the audios and the mel spectrograms:
 `python synthesizer_preprocess_audio.py <datasets_root>`
-
+Allow parameter `--dataset {dataset}` to support adatatang_200zh, SLR68
 * Preprocess the embeddings:
 `python synthesizer_preprocess_embeds.py <datasets_root>/SV2TTS/synthesizer`
 
@@ -56,6 +56,6 @@ or
 
 ## TODO
 - [x] Add demo video
-- [ ] Add support for more dataset
+- [X] Add support for more dataset
 - [ ] Upload pretrained model
 - 🙏 Welcome to add more

synthesizer/preprocess.py

@@ -6,22 +6,28 @@ from pathlib import Path
 from tqdm import tqdm
 import numpy as np
 from encoder import inference as encoder
-from synthesizer.preprocess_speaker import preprocess_speaker_aidatatang_200zh
+from synthesizer.preprocess_speaker import preprocess_speaker_general
 
 data_info = {
     "aidatatang_200zh": {
         "subfolders": ["corpus/train"],
-        "speak_func": preprocess_speaker_aidatatang_200zh
-    }
-    # TODO add more
+        "trans_filepath": "transcript/aidatatang_200_zh_transcript.txt",
+        "speak_func": preprocess_speaker_general
+    },
+    "SLR68": {
+        "subfolders": ["train"],
+        "trans_filepath": "train/TRANS.txt",
+        "speak_func": preprocess_speaker_general
+    },
 }
 
 def preprocess_dataset(datasets_root: Path, out_dir: Path, n_processes: int,
                            skip_existing: bool, hparams, no_alignments: bool,
                            dataset: str):
+    dataset_info = data_info[dataset]
     # Gather the input directories
     dataset_root = datasets_root.joinpath(dataset)
-    input_dirs = [dataset_root.joinpath(subfolder.strip()) for subfolder in data_info[dataset]["subfolders"]]
+    input_dirs = [dataset_root.joinpath(subfolder.strip()) for subfolder in dataset_info["subfolders"]]
     print("\n    ".join(map(str, ["Using data from:"] + input_dirs)))
     assert all(input_dir.exists() for input_dir in input_dirs)
     
@@ -34,9 +40,19 @@ def preprocess_dataset(datasets_root: Path, out_dir: Path, n_processes: int,
     metadata_file = metadata_fpath.open("a" if skip_existing else "w", encoding="utf-8")
 
     # Preprocess the dataset
+    dict_info = {}
+    transcript_dirs = dataset_root.joinpath(dataset_info["trans_filepath"])
+    assert transcript_dirs.exists(), str(transcript_dirs)+" not exist."
+    with open(transcript_dirs, "r", encoding="utf-8") as dict_transcript:
+        for v in dict_transcript:
+            if not v:
+                continue
+            v = v.strip().replace("\n","").replace("\t"," ").split(" ")
+            dict_info[v[0]] = " ".join(v[1:])
+
     speaker_dirs = list(chain.from_iterable(input_dir.glob("*") for input_dir in input_dirs))
-    func = partial(data_info[dataset]["speak_func"], out_dir=out_dir, skip_existing=skip_existing, 
-                   hparams=hparams, directory=dataset_root, no_alignments=no_alignments)
+    func = partial(dataset_info["speak_func"], out_dir=out_dir, skip_existing=skip_existing, 
+                   hparams=hparams, dict_info=dict_info, no_alignments=no_alignments)
     job = Pool(n_processes).imap(func, speaker_dirs)
     for speaker_metadata in tqdm(job, dataset, len(speaker_dirs), unit="speakers"):
         for metadatum in speaker_metadata:

synthesizer/preprocess_speaker.py

@@ -1,4 +1,3 @@
-import platform
 import librosa
 import numpy as np
 
@@ -68,36 +67,30 @@ def _split_on_silences_aidatatang_200zh(wav_fpath, words, hparams):
     wav = librosa.effects.trim(wav, top_db= 40, frame_length=2048, hop_length=512)[0]
     if hparams.rescale:
         wav = wav / np.abs(wav).max() * hparams.rescaling_max
+    # denoise, we may not need it here.
+    if len(wav) > hparams.sample_rate*(0.3+0.1):
+        noise_wav = np.concatenate([wav[:int(hparams.sample_rate*0.15)],
+                                    wav[-int(hparams.sample_rate*0.15):]])
+        profile = logmmse.profile_noise(noise_wav, hparams.sample_rate)
+        wav = logmmse.denoise(wav, profile, eta=0)
+
     resp = pinyin(words, style=Style.TONE3)
     res = [v[0] for v in resp if v[0].strip()]
     res = " ".join(res)
 
     return wav, res
 
-
-def preprocess_speaker_aidatatang_200zh(speaker_dir, out_dir: Path, skip_existing: bool, hparams, directory, no_alignments: bool):
-    dict_info = {}
-    transcript_dirs = directory.joinpath("transcript/aidatatang_200_zh_transcript.txt")
-    with open(transcript_dirs,"rb") as fp:
-        dict_transcript = [v.decode() for v in fp]
-    for v in dict_transcript:
-        if not v:
-            continue
-        v = v.strip().replace("\n","").split(" ")
-        dict_info[v[0]] = " ".join(v[1:])
-
+def preprocess_speaker_general(speaker_dir, out_dir: Path, skip_existing: bool, hparams, dict_info, no_alignments: bool):
     metadata = []
-    if platform.system() == "Windows":
-        split = "\\"
-    else:
-        split = "/" 
-    for wav_fpath in speaker_dir.glob("*.wav"):
-        name = str(wav_fpath).split(split)[-1]
-        key = name.split(".")[0]
-        words = dict_info.get(key)
+    wav_fpath_list = speaker_dir.glob("*.wav")
+    # Iterate over each wav
+    for wav_fpath in wav_fpath_list:
+        words = dict_info.get(wav_fpath.name.split(".")[0])
+        words = dict_info.get(wav_fpath.name) if not words else words # try with wav 
         if not words:
+            print("no wordS")
             continue
-        sub_basename = "%s_%02d" % (name, 0)
+        sub_basename = "%s_%02d" % (wav_fpath.name, 0)
         wav, text = _split_on_silences_aidatatang_200zh(wav_fpath, words, hparams)
         metadata.append(_process_utterance(wav, text, out_dir, sub_basename, 
                                               skip_existing, hparams))