add paddle2onnx, test=tts

examples/csmsc/tts3/local/ort_predict.sh

@@ -3,7 +3,7 @@ train_output_path=$1
 stage=0
 stop_stage=0
 
-# only support default_fastspeech2 + hifigan now!
+# only support default_fastspeech2 + hifigan/mb_melgan now!
 
 # synthesize from metadata
 if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
@@ -28,4 +28,4 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
         --phones_dict=dump/phone_id_map.txt \
         --device=cpu \
         --cpu_threads=2
-fi
+fi

examples/csmsc/tts3/local/paddle2onnx.sh

@@ -0,0 +1,22 @@
+train_output_path=$1
+model_dir=$2
+output_dir=$3
+model=$4
+
+enable_dev_version=True
+
+model_name=${model%_*}
+echo model_name: ${model_name}
+
+if [ ${model_name} = 'mb_melgan' ] ;then
+    enable_dev_version=False
+fi
+
+mkdir -p ${train_output_path}/${output_dir}
+
+paddle2onnx \
+    --model_dir ${train_output_path}/${model_dir} \
+    --model_filename ${model}.pdmodel \
+    --params_filename ${model}.pdiparams \
+    --save_file ${train_output_path}/${output_dir}/${model}.onnx \
+    --enable_dev_version ${enable_dev_version}

examples/csmsc/tts3/run.sh

@@ -41,3 +41,17 @@ if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
     CUDA_VISIBLE_DEVICES=${gpus} ./local/inference.sh ${train_output_path} || exit -1
 fi
 
+# paddle2onnx, please make sure the static models are in ${train_output_path}/inference first
+# we have only tested the following models so far
+if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
+    pip install paddle2onnx==0.9.4
+    ./local/paddle2onnx.sh ${train_output_path} inference inference_onnx fastspeech2_csmsc
+    ./local/paddle2onnx.sh ${train_output_path} inference inference_onnx hifigan_csmsc
+    ./local/paddle2onnx.sh ${train_output_path} inference inference_onnx mb_melgan_csmsc
+fi
+
+# inference with onnxruntime, use fastspeech2 + hifigan by default
+if [ ${stage} -le 6 ] && [ ${stop_stage} -ge 6 ]; then
+    # pip install onnxruntime
+    ./local/ort_predict.sh ${train_output_path}
+fi