Merge pull request #190 from yt605155624/fs2_en

Refactor of examples, add fastspeech2_ljspeech, pwg_ljspeech, update …

.gitignore

@@ -143,4 +143,11 @@ dmypy.json
 runs
 syn_audios
 exp/
+exp_*/
 dump/
+dump_*/
+*_test.py
+*_torch.py
+*.ipynb
+*ckpt*
+*alignment*

benchmark/PWGAN/run_all.sh

@@ -14,18 +14,19 @@ fi
 # 2 拷贝该模型需要数据、预训练模型
 # 下载 baker 数据集到 home 目录下并解压缩到 home 目录下
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
-      wget https://paddlespeech.bj.bcebos.com/datasets/BZNSYP.rar
+      wget https://weixinxcxdb.oss-cn-beijing.aliyuncs.com/gwYinPinKu/BZNSYP.rar
       mkdir BZNSYP
       unrar x BZNSYP.rar BZNSYP
+      wget https://paddlespeech.bj.bcebos.com/Parakeet/benchmark/durations.txt
 fi
 # 数据预处理
 if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
 
-      python examples/parallelwave_gan/baker/preprocess.py --rootdir=BZNSYP/ --dumpdir=dump --num_cpu=20
-      python examples/parallelwave_gan/baker/compute_statistics.py --metadata=dump/train/raw/metadata.jsonl --field-name="feats" --dumpdir=dump/train
-      python examples/parallelwave_gan/baker/normalize.py --metadata=dump/train/raw/metadata.jsonl --dumpdir=dump/train/norm --stats=dump/train/stats.npy
-      python examples/parallelwave_gan/baker/normalize.py --metadata=dump/dev/raw/metadata.jsonl --dumpdir=dump/dev/norm --stats=dump/train/stats.npy
-      python examples/parallelwave_gan/baker/normalize.py --metadata=dump/test/raw/metadata.jsonl --dumpdir=dump/test/norm --stats=dump/train/stats.npy
+      python utils/vocoder_preprocess.py --rootdir=BZNSYP/ --dumpdir=dump --num-cpu=20 --cut-sil=True --dur-file=durations.txt --config=examples/parallelwave_gan/baker/conf/default.yaml
+      python utils/compute_statistics.py --metadata=dump/train/raw/metadata.jsonl --field-name="feats"
+      python utils/vocoder_normalize.py --metadata=dump/train/raw/metadata.jsonl --dumpdir=dump/train/norm --stats=dump/train/feats_stats.npy
+      python utils/vocoder_normalize.py --metadata=dump/dev/raw/metadata.jsonl --dumpdir=dump/dev/norm --stats=dump/train/feats_stats.npy
+      python utils/vocoder_normalize.py --metadata=dump/test/raw/metadata.jsonl --dumpdir=dump/test/norm --stats=dump/train/feats_stats.npy
 fi
 # 3 批量运行（如不方便批量，1，2需放到单个模型中）
 if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then

benchmark/PWGAN/run_benchmark.sh

@@ -29,8 +29,8 @@ function _train(){
                --run-benchmark=true"   
 
     case ${run_mode} in
-    sp) train_cmd="python3 examples/parallelwave_gan/baker/train.py --nprocs=1 ${train_cmd}" ;;
-    mp) train_cmd="python3 examples/parallelwave_gan/baker/train.py --nprocs=8 ${train_cmd}"
+    sp) train_cmd="python3 utils/pwg_train.py --nprocs=1 ${train_cmd}" ;;
+    mp) train_cmd="python3 utils/pwg_train.py --nprocs=8 ${train_cmd}"
         log_parse_file="mylog/workerlog.0" ;;
     *) echo "choose run_mode(sp or mp)"; exit 1;
     esac

examples/fastspeech2/aishell3/README.md

@@ -15,7 +15,7 @@ Extract AISHELL-3.
 mkdir data_aishell3
 tar zxvf data_aishell3.tgz -C data_aishell3
 ```
-### Get MFA result of BZNSYP and Extract it
+### Get MFA result of AISHELL-3 and Extract it
 We use [MFA2.x](https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner) to get durations for aishell3_fastspeech2.
 You can download from here [aishell3_alignment_tone.tar.gz](https://paddlespeech.bj.bcebos.com/MFA/AISHELL-3/with_tone/aishell3_alignment_tone.tar.gz), or train your own MFA model reference to [use_mfa example](https://github.com/PaddlePaddle/Parakeet/tree/develop/examples/use_mfa) (use MFA1.x now) of our repo.
 ### Preprocess the dataset
@@ -48,21 +48,25 @@ The dataset is split into 3 parts, namely `train`, `dev` and` test`, each of whi
 Also there is a `metadata.jsonl` in each subfolder. It is a table-like file which contains phones, text_lengths, speech_lengths, durations, path of speech features, path of pitch features, path of energy features, speaker and id of each utterance.
 
 ## Train the model
+`./run.sh` calls `Parakeet/utils/multi_spk_fs2_train.py`.
 ```bash
 ./run.sh
 ```
-Or you can use `train.py` directly. Here's the complete help message.
+Here's the complete help message.
 ```text
-usage: train.py [-h] [--config CONFIG] [--train-metadata TRAIN_METADATA]
-                [--dev-metadata DEV_METADATA] [--output-dir OUTPUT_DIR]
-                [--device DEVICE] [--nprocs NPROCS] [--verbose VERBOSE]
-                [--phones-dict PHONES_DICT] [--speaker-dict SPEAKER_DICT]
+usage: multi_spk_fs2_train.py [-h] [--config CONFIG]
+                              [--train-metadata TRAIN_METADATA]
+                              [--dev-metadata DEV_METADATA]
+                              [--output-dir OUTPUT_DIR] [--device DEVICE]
+                              [--nprocs NPROCS] [--verbose VERBOSE]
+                              [--phones-dict PHONES_DICT]
+                              [--speaker-dict SPEAKER_DICT]
 
-Train a FastSpeech2 model with AISHELL-3 Mandrin TTS dataset.
+Train a FastSpeech2 model with multiple speaker dataset.
 
 optional arguments:
   -h, --help            show this help message and exit
-  --config CONFIG       config file to overwrite default config.
+  --config CONFIG       fastspeech2 config file.
   --train-metadata TRAIN_METADATA
                         training data.
   --dev-metadata DEV_METADATA

examples/fastspeech2/aishell3/preprocess.sh

@@ -20,12 +20,12 @@ fi
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     # extract features
     echo "Extract features ..."
-    python3 ${MAIN_ROOT}/utils/fastspeech2_preprocess.py \
+    python3 ${MAIN_ROOT}/utils/fs2_preprocess.py \
         --dataset=aishell3 \
         --rootdir=~/datasets/data_aishell3/ \
         --dumpdir=dump \
         --dur-file=durations.txt \
-        --config-path=conf/default.yaml \
+        --config=conf/default.yaml \
         --num-cpu=8 \
         --cut-sil=True
 fi
@@ -47,9 +47,9 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
 fi
 
 if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
-    # normalize and covert phone to id, dev and test should use train's stats
+    # normalize and covert phone/speaker to id, dev and test should use train's stats
     echo "Normalize ..."
-    python3 ${MAIN_ROOT}/utils/fastspeech2_normalize.py \
+    python3 ${MAIN_ROOT}/utils/fs2_normalize.py \
         --metadata=dump/train/raw/metadata.jsonl \
         --dumpdir=dump/train/norm \
         --speech-stats=dump/train/speech_stats.npy \
@@ -58,7 +58,7 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
         --phones-dict=dump/phone_id_map.txt \
         --speaker-dict=dump/speaker_id_map.txt
 
-    python3 ${MAIN_ROOT}/utils/fastspeech2_normalize.py \
+    python3 ${MAIN_ROOT}/utils/fs2_normalize.py \
         --metadata=dump/dev/raw/metadata.jsonl \
         --dumpdir=dump/dev/norm \
         --speech-stats=dump/train/speech_stats.npy \
@@ -67,7 +67,7 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
         --phones-dict=dump/phone_id_map.txt \
         --speaker-dict=dump/speaker_id_map.txt
 
-    python3 ${MAIN_ROOT}/utils/fastspeech2_normalize.py \
+    python3 ${MAIN_ROOT}/utils/fs2_normalize.py \
         --metadata=dump/test/raw/metadata.jsonl \
         --dumpdir=dump/test/norm \
         --speech-stats=dump/train/speech_stats.npy \

examples/fastspeech2/aishell3/run.sh

@@ -1,6 +1,8 @@
 #!/bin/bash
 
-python3 train.py \
+export MAIN_ROOT=`realpath ${PWD}/../../../`
+
+python3 ${MAIN_ROOT}/utils/multi_spk_fs2_train.py \
     --train-metadata=dump/train/norm/metadata.jsonl \
     --dev-metadata=dump/dev/norm/metadata.jsonl \
     --config=conf/default.yaml \

examples/fastspeech2/aishell3/synthesize.sh

@@ -1,6 +1,8 @@
 #!/bin/bash
 
-python3 synthesize.py \
+export MAIN_ROOT=`realpath ${PWD}/../../../`
+
+python3  ${MAIN_ROOT}/utils/multi_spk_fs2_pwg_syn.py \
   --fastspeech2-config=conf/default.yaml \
   --fastspeech2-checkpoint=exp/default/checkpoints/snapshot_iter_96400.pdz \
   --fastspeech2-stat=dump/train/speech_stats.npy \

examples/fastspeech2/aishell3/synthesize_e2e.py

@@ -20,14 +20,13 @@
 import paddle
 import soundfile as sf
 import yaml
-from yacs.config import CfgNode
+from parakeet.frontend.cn_frontend import Frontend
 from parakeet.models.fastspeech2 import FastSpeech2
 from parakeet.models.fastspeech2 import FastSpeech2Inference
 from parakeet.models.parallel_wavegan import PWGGenerator
 from parakeet.models.parallel_wavegan import PWGInference
 from parakeet.modules.normalizer import ZScore
-
-from frontend import Frontend
+from yacs.config import CfgNode
 
 
 def evaluate(args, fastspeech2_config, pwg_config):
@@ -67,7 +66,7 @@ def evaluate(args, fastspeech2_config, pwg_config):
     vocoder.eval()
     print("model done!")
 
-    frontend = Frontend(args.phones_dict)
+    frontend = Frontend(phone_vocab_path=args.phones_dict)
     print("frontend done!")
 
     stat = np.load(args.fastspeech2_stat)