add with no grad when inference

PaddlePaddle · Oct 12, 2021 · aacf81c · aacf81c
1 parent 4b94af9
commit aacf81c
Show file tree

Hide file tree

Showing 7 changed files with 25 additions and 23 deletions.
diff --git a/examples/GANVocoder/parallelwave_gan/baker/synthesize_from_wav.py b/examples/GANVocoder/parallelwave_gan/baker/synthesize_from_wav.py
@@ -76,7 +76,8 @@ def evaluate(args, config):
         # extract mel feats
         mel = mel_extractor.get_log_mel_fbank(wav)
         mel = paddle.to_tensor(mel)
-        gen_wav = pwg_inference(mel)
+        with paddle.no_grad():
+            gen_wav = pwg_inference(mel)
         sf.write(
             str(output_dir / ("gen_" + utt_name)),
             gen_wav.numpy(),

diff --git a/examples/GANVocoder/parallelwave_gan/synthesize.py b/examples/GANVocoder/parallelwave_gan/synthesize.py
@@ -80,7 +80,8 @@ def main():
         mel = example['feats']
         mel = paddle.to_tensor(mel)  # (T, C)
         with timer() as t:
-            wav = generator.inference(c=mel)
+            with paddle.no_grad():
+                wav = generator.inference(c=mel)
             wav = wav.numpy()
             N += wav.size
             T += t.elapse

diff --git a/examples/fastspeech2/aishell3/synthesize_e2e.py b/examples/fastspeech2/aishell3/synthesize_e2e.py
@@ -97,11 +97,11 @@ def evaluate(args, fastspeech2_config, pwg_config):
                 mel = fastspeech2_inference(
                     part_phone_ids, spk_id=paddle.to_tensor(spk_id))
                 temp_wav = pwg_inference(mel)
-                if flags == 0:
-                    wav = temp_wav
-                    flags = 1
-                else:
-                    wav = paddle.concat([wav, temp_wav])
+            if flags == 0:
+                wav = temp_wav
+                flags = 1
+            else:
+                wav = paddle.concat([wav, temp_wav])
         sf.write(
             str(output_dir / (str(spk_id) + "_" + utt_id + ".wav")),
             wav.numpy(),

diff --git a/examples/fastspeech2/baker/synthesize_e2e.py b/examples/fastspeech2/baker/synthesize_e2e.py
@@ -87,11 +87,11 @@ def evaluate(args, fastspeech2_config, pwg_config):
             with paddle.no_grad():
                 mel = fastspeech2_inference(part_phone_ids)
                 temp_wav = pwg_inference(mel)
-                if flags == 0:
-                    wav = temp_wav
-                    flags = 1
-                else:
-                    wav = paddle.concat([wav, temp_wav])
+            if flags == 0:
+                wav = temp_wav
+                flags = 1
+            else:
+                wav = paddle.concat([wav, temp_wav])
         sf.write(
             str(output_dir / (utt_id + ".wav")),
             wav.numpy(),

diff --git a/examples/speedyspeech/baker/synthesize_e2e.py b/examples/speedyspeech/baker/synthesize_e2e.py
@@ -121,11 +121,11 @@ def evaluate(args, speedyspeech_config, pwg_config):
             with paddle.no_grad():
                 mel = speedyspeech_inference(part_phone_ids, part_tone_ids)
                 temp_wav = pwg_inference(mel)
-                if flags == 0:
-                    wav = temp_wav
-                    flags = 1
-                else:
-                    wav = paddle.concat([wav, temp_wav])
+            if flags == 0:
+                wav = temp_wav
+                flags = 1
+            else:
+                wav = paddle.concat([wav, temp_wav])
         sf.write(
             output_dir / (utt_id + ".wav"),
             wav.numpy(),

diff --git a/examples/transformer_tts/ljspeech/synthesize_e2e.py b/examples/transformer_tts/ljspeech/synthesize_e2e.py
@@ -89,11 +89,12 @@ def evaluate(args, acoustic_model_config, vocoder_config):
         phones = [phn for phn in phones if not phn.isspace()]
         phones = [phn if phn in phone_id_map else "," for phn in phones]
         phone_ids = [phone_id_map[phn] for phn in phones]
-        mel = transformer_tts_inference(paddle.to_tensor(phone_ids))
-        # mel shape is (T, feats) and waveflow's input shape is (batch, feats, T)
-        mel = mel.unsqueeze(0).transpose([0, 2, 1])
-        # wavflow's output shape is (B, T)
-        wav = vocoder.infer(mel)[0]
+        with paddle.no_grad():
+            mel = transformer_tts_inference(paddle.to_tensor(phone_ids))
+            # mel shape is (T, feats) and waveflow's input shape is (batch, feats, T)
+            mel = mel.unsqueeze(0).transpose([0, 2, 1])
+            # wavflow's output shape is (B, T)
+            wav = vocoder.infer(mel)[0]
 
         sf.write(
             str(output_dir / (utt_id + ".wav")),

diff --git a/examples/transformer_tts/synthesize.py b/examples/transformer_tts/synthesize.py
@@ -81,7 +81,6 @@ def evaluate(args, acoustic_model_config, vocoder_config):
             mel = mel.unsqueeze(0).transpose([0, 2, 1])
             # wavflow's output shape is (B, T)
             wav = vocoder.infer(mel)[0]
-            print("wav:", wav)
 
         sf.write(
             str(output_dir / (utt_id + ".wav")),