add docker support (#802)

* add docker support

* 修复训练集解压问题

* 修复web.py启动问题

* 混合数据集训练参数

.dockerignore

@@ -0,0 +1,4 @@
+*/saved_models
+!vocoder/saved_models/pretrained/**
+!encoder/saved_models/pretrained.pt
+/datasets

.gitignore

@@ -19,3 +19,6 @@
 !encoder/saved_models/pretrained.pt
 wavs
 log
+!/docker-entrypoint.sh
+!/datasets_download/*.sh
+/datasets

Dockerfile

@@ -0,0 +1,17 @@
+FROM pytorch/pytorch:latest
+
+RUN apt-get update && apt-get install -y build-essential ffmpeg parallel aria2 && apt-get clean
+
+COPY ./requirements.txt /workspace/requirements.txt
+
+RUN pip install -r requirements.txt && pip install webrtcvad-wheels
+
+COPY . /workspace
+
+VOLUME [ "/datasets", "/workspace/synthesizer/saved_models/" ]
+
+ENV DATASET_MIRROR=default FORCE_RETRAIN=false TRAIN_DATASETS=aidatatang_200zh\ magicdata\ aishell3\ data_aishell TRAIN_SKIP_EXISTING=true
+
+EXPOSE 8080
+
+ENTRYPOINT [ "/workspace/docker-entrypoint.sh" ]

datasets_download/CN.txt

@@ -0,0 +1,8 @@
+https://openslr.magicdatatech.com/resources/62/aidatatang_200zh.tgz
+  out=download/aidatatang_200zh.tgz
+https://openslr.magicdatatech.com/resources/68/train_set.tar.gz
+  out=download/magicdata.tgz
+https://openslr.magicdatatech.com/resources/93/data_aishell3.tgz
+  out=download/aishell3.tgz
+https://openslr.magicdatatech.com/resources/33/data_aishell.tgz
+  out=download/data_aishell.tgz

datasets_download/EU.txt

@@ -0,0 +1,8 @@
+https://openslr.elda.org/resources/62/aidatatang_200zh.tgz
+  out=download/aidatatang_200zh.tgz
+https://openslr.elda.org/resources/68/train_set.tar.gz
+  out=download/magicdata.tgz
+https://openslr.elda.org/resources/93/data_aishell3.tgz
+  out=download/aishell3.tgz
+https://openslr.elda.org/resources/33/data_aishell.tgz
+  out=download/data_aishell.tgz

datasets_download/US.txt

@@ -0,0 +1,8 @@
+https://us.openslr.org/resources/62/aidatatang_200zh.tgz
+  out=download/aidatatang_200zh.tgz
+https://us.openslr.org/resources/68/train_set.tar.gz
+  out=download/magicdata.tgz
+https://us.openslr.org/resources/93/data_aishell3.tgz
+  out=download/aishell3.tgz
+https://us.openslr.org/resources/33/data_aishell.tgz
+  out=download/data_aishell.tgz

datasets_download/datasets.sha256sum

@@ -0,0 +1,4 @@
+0c0ace77fe8ee77db8d7542d6eb0b7ddf09b1bfb880eb93a7fbdbf4611e9984b  /datasets/download/aidatatang_200zh.tgz
+be2507d431ad59419ec871e60674caedb2b585f84ffa01fe359784686db0e0cc  /datasets/download/aishell3.tgz
+a4a0313cde0a933e0e01a451f77de0a23d6c942f4694af5bb7f40b9dc38143fe  /datasets/download/data_aishell.tgz
+1d2647c614b74048cfe16492570cc5146d800afdc07483a43b31809772632143  /datasets/download/magicdata.tgz

datasets_download/default.txt

@@ -0,0 +1,8 @@
+https://www.openslr.org/resources/62/aidatatang_200zh.tgz
+  out=download/aidatatang_200zh.tgz
+https://www.openslr.org/resources/68/train_set.tar.gz
+  out=download/magicdata.tgz
+https://www.openslr.org/resources/93/data_aishell3.tgz
+  out=download/aishell3.tgz
+https://www.openslr.org/resources/33/data_aishell.tgz
+  out=download/data_aishell.tgz

datasets_download/download.sh

@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+
+set -Eeuo pipefail
+
+aria2c -x 10 --disable-ipv6 --input-file /workspace/datasets_download/${DATASET_MIRROR}.txt --dir /datasets --continue
+
+echo "Verifying sha256sum..."
+parallel --will-cite -a /workspace/datasets_download/datasets.sha256sum "echo -n {} | sha256sum -c"

datasets_download/extract.sh

@@ -0,0 +1,29 @@
+#!/usr/bin/env bash
+
+set -Eeuo pipefail
+
+mkdir -p /datasets/aidatatang_200zh
+if [ -z "$(ls -A /datasets/aidatatang_200zh)" ] ; then
+    tar xvz --directory /datasets/ -f /datasets/download/aidatatang_200zh.tgz --exclude 'aidatatang_200zh/corpus/dev/*' --exclude 'aidatatang_200zh/corpus/test/*'
+    cd /datasets/aidatatang_200zh/corpus/train/
+    cat *.tar.gz | tar zxvf - -i
+    rm -f *.tar.gz
+fi
+
+mkdir -p /datasets/magicdata
+if [ -z "$(ls -A /datasets/magicdata)" ] ; then
+    tar xvz --directory /datasets/magicdata -f /datasets/download/magicdata.tgz train/
+fi
+
+mkdir -p /datasets/aishell3
+if [ -z "$(ls -A /datasets/aishell3)" ] ; then
+    tar xvz --directory /datasets/aishell3 -f /datasets/download/aishell3.tgz train/
+fi
+
+mkdir -p /datasets/data_aishell
+if [ -z "$(ls -A /datasets/data_aishell)" ] ; then
+    tar xvz --directory /datasets/ -f /datasets/download/data_aishell.tgz
+    cd /datasets/data_aishell/wav/
+    cat *.tar.gz | tar zxvf - -i --exclude 'dev/*' --exclude 'test/*'
+    rm -f *.tar.gz
+fi

docker-compose.yml

@@ -0,0 +1,23 @@
+version: '3.8'
+
+services:
+  server:
+    image: mockingbird:latest
+    build: .
+    volumes:
+      - ./datasets:/datasets
+      - ./synthesizer/saved_models:/workspace/synthesizer/saved_models
+    environment:
+      - DATASET_MIRROR=US
+      - FORCE_RETRAIN=false
+      - TRAIN_DATASETS=aidatatang_200zh magicdata aishell3 data_aishell
+      - TRAIN_SKIP_EXISTING=true
+    ports:
+      - 8080:8080
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              device_ids: [ '0' ]
+              capabilities: [ gpu ]

docker-entrypoint.sh

@@ -0,0 +1,17 @@
+#!/usr/bin/env bash
+
+if [ -z "$(ls -A /workspace/synthesizer/saved_models)" ] || [ "$FORCE_RETRAIN" = true ] ; then
+    /workspace/datasets_download/download.sh
+    /workspace/datasets_download/extract.sh
+    for DATASET in ${TRAIN_DATASETS}
+    do
+        if [ "$TRAIN_SKIP_EXISTING" = true ] ; then
+            python pre.py /datasets -d ${DATASET} -n $(nproc) --skip_existing
+        else
+            python pre.py /datasets -d ${DATASET} -n $(nproc)
+        fi
+    done
+    python synthesizer_train.py mandarin /datasets/SV2TTS/synthesizer
+fi
+
+python web.py

requirements.txt

@@ -26,3 +26,7 @@ PyYAML==5.4.1
 torch_complex
 espnet
 PyWavelets
+fastapi
+loguru
+typer[all]
+click==8.0.4