[TTS]Fix VITS lite infer (#3098)

examples/csmsc/vits/local/lite_predict.sh

@@ -7,7 +7,7 @@ stage=0
 stop_stage=0
 
 if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
-    python3 ${BIN_DIR}/../lite_predict.py \
+    python3 ${BIN_DIR}/lite_predict.py \
         --inference_dir=${train_output_path}/pdlite \
         --am=vits_csmsc \
         --text=${BIN_DIR}/../sentences.txt \

examples/csmsc/vits/run.sh

@@ -54,16 +54,16 @@ fi
 #     ./local/ort_predict.sh ${train_output_path}
 # fi
 
-# # not ready yet for operator missing in Paddle-Lite
+# not ready yet for operator missing in Paddle-Lite
-# # must run after stage 3 (which stage generated static models)
+# must run after stage 3 (which stage generated static models)
-# if [ ${stage} -le 7 ] && [ ${stop_stage} -ge 7 ]; then
+if [ ${stage} -le 7 ] && [ ${stop_stage} -ge 7 ]; then
-#     # NOTE by yuantian 2022.11.21: please compile develop version of Paddle-Lite to export and run TTS models,
+    # NOTE by yuantian 2022.11.21: please compile develop version of Paddle-Lite to export and run TTS models,
-#     #                   cause TTS models are supported by https://github.com/PaddlePaddle/Paddle-Lite/pull/9587 
+    #                   cause TTS models are supported by https://github.com/PaddlePaddle/Paddle-Lite/pull/10128
-#     #                   and https://github.com/PaddlePaddle/Paddle-Lite/pull/9706
+    # vits can only run in arm
-#     ./local/export2lite.sh ${train_output_path} inference pdlite vits_csmsc x86
+    ./local/export2lite.sh ${train_output_path} inference pdlite vits_csmsc arm
-# fi
+fi
 
-# if [ ${stage} -le 8 ] && [ ${stop_stage} -ge 8 ]; then
+if [ ${stage} -le 8 ] && [ ${stop_stage} -ge 8 ]; then
-#     CUDA_VISIBLE_DEVICES=${gpus} ./local/lite_predict.sh ${train_output_path} || exit -1
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/lite_predict.sh ${train_output_path} || exit -1
-# fi
+fi
 

paddlespeech/t2s/exps/vits/lite_predict.py

@@ -21,6 +21,7 @@ from paddlespeech.t2s.exps.lite_syn_utils import get_lite_am_output
 from paddlespeech.t2s.exps.lite_syn_utils import get_lite_predictor
 from paddlespeech.t2s.exps.syn_utils import get_frontend
 from paddlespeech.t2s.exps.syn_utils import get_sentences
+from paddlespeech.t2s.utils import str2bool
 
 
 def parse_args():
@@ -75,12 +76,12 @@ def main():
     # frontend
     frontend = get_frontend(
         lang=args.lang,
-        phones_dict=args.phones_dict,
+        phones_dict=args.phones_dict)
-        tones_dict=args.tones_dict)
 
     # am_predictor
+    # vits can only run in arm
     am_predictor = get_lite_predictor(
-        model_dir=args.inference_dir, model_file=args.am + "_x86.nb")
+        model_dir=args.inference_dir, model_file=args.am + "_arm.nb")
     # model: {model_name}_{dataset}
     am_dataset = args.am[args.am.rindex('_') + 1:]