From 88f4369c636060ce7c7ee0740d631259bda6f424 Mon Sep 17 00:00:00 2001 From: Edresson Date: Mon, 22 Nov 2021 18:43:56 -0300 Subject: [PATCH] Fix the bug in multispeaker vits --- TTS/bin/train_tts.py | 9 +++++++++ tests/tts_tests/test_vits_multilingual_train.py | 6 ++++++ 2 files changed, 15 insertions(+) diff --git a/TTS/bin/train_tts.py b/TTS/bin/train_tts.py index e28e9dec..f39ed259 100644 --- a/TTS/bin/train_tts.py +++ b/TTS/bin/train_tts.py @@ -47,8 +47,17 @@ def main(): # init speaker manager if config.use_speaker_embedding: speaker_manager = SpeakerManager(data_items=train_samples + eval_samples) + if hasattr(config, "model_args"): + config.model_args.num_speakers = len(speaker_manager.speaker_ids) + else: + config.num_speakers = len(speaker_manager.speaker_ids) + elif config.use_d_vector_file: speaker_manager = SpeakerManager(d_vectors_file_path=config.d_vector_file) + if hasattr(config, "model_args"): + config.model_args.num_speakers = len(speaker_manager.speaker_ids) + else: + config.num_speakers = len(speaker_manager.speaker_ids) else: speaker_manager = None diff --git a/tests/tts_tests/test_vits_multilingual_train.py b/tests/tts_tests/test_vits_multilingual_train.py index 04b42e61..a280e8c5 100644 --- a/tests/tts_tests/test_vits_multilingual_train.py +++ b/tests/tts_tests/test_vits_multilingual_train.py @@ -53,9 +53,15 @@ config.audio.trim_db = 60 # active multilingual mode config.model_args.use_language_embedding = True +config.use_language_embedding = True # active multispeaker mode config.model_args.use_speaker_embedding = True +config.use_speaker_embedding = True +# config.num_speakers=1 +# config.model_args.num_speakers=1 + config.model_args.use_d_vector_file = False +config.use_d_vector_file = False # active language sampler config.use_language_weighted_sampler = True