[BUG]: 'NS2Trainer' object has no attribute '_count_parameters'

[a897456]

Amphion/models/tts/naturalspeech2/ns2_trainer.py

Lines 134 to 136 in d335514

	self.logger.info(
	f"Model parameters: {self._count_parameters(self.model)/1e6:.2f}M"
	)

Traceback (most recent call last):
File "E:\00\Amphion-main_old\bins\tts\train.py", line 130, in
main()
File "E:\00\Amphion-main_old\bins\tts\train.py", line 104, in main
trainer = build_trainer(args, cfg)
File "E:\00\Amphion-main_old\bins\tts\train.py", line 26, in build_trainer
trainer = trainer_class(args, cfg)#NS2Trainer
File "E:\00\Amphion-main_old\models\tts\naturalspeech2\ns2_trainer.py", line 135, in init
f"Model parameters: {self._count_parameters(self.model)/1e6:.2f}M"
AttributeError: 'NS2Trainer' object has no attribute '_count_parameters'

[a897456]

Amphion/egs/tts/NaturalSpeech2/exp_config.json

Line 11 in d335514

"read_metadata": true,

I think it is false.

[netagl]

I changes the __count_parameters(model) in TTSTrainer class, to _count_parameters(model)
@a897456

[a897456]

I changes the __count_parameters(model) in TTSTrainer class, to _count_parameters(model) @a897456

Yes, _dump_cfg is also.
And do you ever met : FileNotFoundError: [Errno 2] No such file or directory: 'data\libritts\code\19\train-clean-100#19#198#19_198_000004_000000.npy' in there:

Amphion/models/tts/naturalspeech2/ns2_dataset.py

Lines 187 to 196 in d335514

	if self.cfg.preprocess.read_metadata:
	metadata_uid_path = os.path.join(
	self.cfg.preprocess.processed_dir,
	self.cfg.preprocess.metadata_dir,
	dataset,
	# utt_info["speaker"],
	uid + ".pkl",
	)
	with open(metadata_uid_path, "rb") as f:
	metadata_uid = pickle.load(f)

[netagl]

yes, also _dump_cfg.

yeah. @a897456
I changed self.cfg.preprocess.read_metadata: to False, and used acoustic_extractor to create this files:

            # code
            code = np.load(self.utt2code_path[utt])
            # frame_nums
            frame_nums = code.shape[1]
            # pitch
            pitch = np.load(self.utt2pitch_path[utt])
            # duration
            duration = np.load(self.utt2duration_path[utt])
            # phone_id
            phone_id = np.array(
                [
                    *map(
                        self.phone2id.get,
                        self.utt2phone[utt].replace("{", "").replace("}", "").split(),
                    )
                ]
            )

[a897456]

yeah. @a897456 I changed self.cfg.preprocess.read_metadata: to False, and used acoustic_extractor to create this files:

used acoustic_extractor to create this files？
How?

            # code
            code = np.load(self.utt2code_path[utt])
            # frame_nums
            frame_nums = code.shape[1]
            # pitch
            pitch = np.load(self.utt2pitch_path[utt])
            # duration
            duration = np.load(self.utt2duration_path[utt])
            # phone_id
            phone_id = np.array(
                [
                    *map(
                        self.phone2id.get,
                        self.utt2phone[utt].replace("{", "").replace("}", "").split(),
                    )
                ]
            )

This is the code in which part of if self.cfg.preprocess.read_metadata is false, so can you show the code of how use acoustic_extractor to create this files?

[netagl]

code:

        if cfg.preprocess.extract_acoustic_token:
            print('extract_acoustic_token')
            if cfg.preprocess.acoustic_token_extractor == "Encodec":
                codes = extract_encodec_token(wav_path)
                save_feature(
                    dataset_output, cfg.preprocess.acoustic_token_dir, uid, codes
                )

pitch:

        if cfg.preprocess.extract_pitch:
            pitch = f0.get_f0(wav, cfg.preprocess)
            save_feature(dataset_output, cfg.preprocess.pitch_dir, uid, pitch)

            if cfg.preprocess.extract_uv:
                assert isinstance(pitch, np.ndarray)
                uv = pitch != 0
                save_feature(dataset_output, cfg.preprocess.uv_dir, uid, uv)

phones:

from g2p_en import G2p
preprocess_english(res["Text"], lexicon, g2p)

@a897456

[a897456]

THS,but :AttributeError: 'list' object has no attribute 'replace'

Amphion/models/tts/naturalspeech2/ns2_dataset.py

Lines 224 to 230 in d335514

	# phone_id
	phone_id = np.array(
	[
	*map(
	self.phone2id.get,
	self.utt2phone[utt].replace("{", "").replace("}", "").split(),
	)

[netaglazer]

Set True in cfg
@a897456

      assert cfg.preprocess.use_phone == True
        if cfg.preprocess.use_phone:
            self.utt2phone = {}
            for utt_info in self.metadata:
                dataset = utt_info["Dataset"]
                uid = utt_info["Uid"]
                utt = "{}_{}".format(dataset, uid)
                self.utt2phone[utt] = utt_info["phones"]

[a897456]

Set True in cfg @a897456

      assert cfg.preprocess.use_phone == True
        if cfg.preprocess.use_phone:
            self.utt2phone = {}
            for utt_info in self.metadata:
                dataset = utt_info["Dataset"]
                uid = utt_info["Uid"]
                utt = "{}_{}".format(dataset, uid)
                self.utt2phone[utt] = utt_info["phones"]

Yes, and I changed the phone_id =...
phone_id = np.array( [ *map( self.phone2id.get, self.utt2phone[utt].replace("{", "").replace("}", "").split(), ) ] )

[a897456]

Amphion/models/tts/naturalspeech2/ns2_dataset.py

Lines 308 to 313 in d335514

	phone_nums = len(phone_id)
	clip_phone_nums = np.random.randint(
	int(phone_nums * 0.1), int(phone_nums * 0.5) + 1
	)
	clip_phone_nums = max(clip_phone_nums, 1)
	assert clip_phone_nums < phone_nums and clip_phone_nums >= 1

in this code:
phone_nums =len(phone_id)=len(tensor(1,X))=1, so phone_nums always=1
because: phone_id = torch.from_numpy(phone_id).unsqueeze(0)
so clip_phone_nums=1
but assert clip_phone_nums < phone_nums and clip_phone_nums >= 1
How to solve it,please?

[a897456]

Amphion/models/tts/naturalspeech2/ns2_dataset.py

Lines 308 to 313 in d335514

	phone_nums = len(phone_id)
	clip_phone_nums = np.random.randint(
	int(phone_nums * 0.1), int(phone_nums * 0.5) + 1
	)
	clip_phone_nums = max(clip_phone_nums, 1)
	assert clip_phone_nums < phone_nums and clip_phone_nums >= 1

in this code:
phone_nums =len(phone_id)=len(tensor(1,X))=1, so phone_nums always=1
because: phone_id = torch.from_numpy(phone_id).unsqueeze(0)
so clip_phone_nums=1
but assert clip_phone_nums < phone_nums and clip_phone_nums >= 1
How to solve it,please?

[CreepJoye]

Hi,@a897456
I meet the same problem and can't fix it,have you solved the problem?
Any advice will be appreciated!!!

[a897456]

Hi,@a897456 I meet the same problem and can't fix it,have you solved the problem? Any advice will be appreciated!!!

我在群里看到你问了，这个BUG作者应该还没修复。

[CreepJoye]

Hi,@a897456 I meet the same problem and can't fix it,have you solved the problem? Any advice will be appreciated!!!

我在群里看到你问了，这个BUG作者应该还没修复。
方便的话可以在群里加个微信吗，想交流学习一下