wassname/TTS
1
0
Fork 0
mirror of https://github.com/wassname/TTS.git synced 2026-09-09 11:16:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

config, benchmark notebook, synthesis fixed

Browse Source
This commit is contained in:
Thomas Werkmeister
2019-06-26 13:31:16 +02:00
parent d172a3d3d5
commit 05ff8801d1
7 changed files with 216 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files
utils/synthesis.py
+7 -6
View File
@@ -35,17 +35,17 @@ def compute_style_mel(style_wav, ap, use_cuda):
return style_mel
def run_model(model, inputs, CONFIG, truncated, style_mel=None):
def run_model(model, inputs, speaker_id, CONFIG, truncated, style_mel=None):
if CONFIG.model == "TacotronGST" and style_mel is not None:
decoder_output, postnet_output, alignments, stop_tokens = model.inference(
inputs, style_mel)
inputs, style_mel, speaker_id)
else:
if truncated:
decoder_output, postnet_output, alignments, stop_tokens = model.inference_truncated(
inputs)
inputs, speaker_id)
else:
decoder_output, postnet_output, alignments, stop_tokens = model.inference(
inputs)
inputs, speaker_id)
return decoder_output, postnet_output, alignments, stop_tokens
@@ -100,12 +100,13 @@ def synthesis(model,
style_mel = compute_style_mel(style_wav, ap, use_cuda)
# preprocess the given text
inputs = text_to_seqvec(text, CONFIG, use_cuda)
speaker_id = speaker_id_var = torch.from_numpy(speaker_id).unsqueeze(0)
speaker_id = np.asarray(speaker_id)
speaker_id = torch.from_numpy(speaker_id).unsqueeze(0)
if use_cuda:
speaker_id.cuda()
# synthesize voice
decoder_output, postnet_output, alignments, stop_tokens = run_model(
model, inputs, CONFIG, truncated, style_mel)
model, inputs, speaker_id, CONFIG, truncated, style_mel)
# convert outputs to numpy
postnet_output, decoder_output, alignment = parse_outputs(
postnet_output, decoder_output, alignments)