[Bug] Input text cannot be None in attn = generate_path

[cod3r0k]

Describe the bug

When I want to train multispeaker VITs model, after epoch 2, it occurs an error as below:

`
....
--> TIME: 2025-01-18 05:51:44 -- STEP: 1141/1144 -- GLOBAL_STEP: 228275
| > loss_disc: 2.55525541305542 (2.3107323108484175)
| > loss_disc_real_0: 0.07618683576583862 (0.1314041224111602)
| > loss_disc_real_1: 0.2294345200061798 (0.18530447144423717)
| > loss_disc_real_2: 0.14496399462223053 (0.21591699230472428)
| > loss_disc_real_3: 0.2127830535173416 (0.2248999754539924)
| > loss_disc_real_4: 0.17633846402168274 (0.2232837798131024)
| > loss_disc_real_5: 0.1818462312221527 (0.21921113943134243)
| > loss_0: 2.55525541305542 (2.3107323108484175)
| > grad_norm_0: tensor(19.7260, device='cuda:0') (tensor(25.7604, device='cuda:0'))
| > loss_gen: 2.512892723083496 (2.6231823042753417)
| > loss_kl: 2.0453853607177734 (2.0651548769680894)
| > loss_feat: 7.6552958488464355 (8.288802446972165)
| > loss_mel: 22.356651306152344 (22.661884306189645)
| > loss_duration: 1.5326560735702515 (1.469221028108329)
| > amp_scaler: 128.0 (160.98159509202455)
| > loss_1: 36.10288619995117 (37.10824498536824)
| > grad_norm_1: tensor(128.5965, device='cuda:0') (tensor(290.4952, device='cuda:0'))
| > current_lr_0: 0.0001997002061640866
| > current_lr_1: 0.0001997002061640866
| > step_time: 3.5176 (2.388011707745551)
| > loader_time: 0.021 (0.007880230309355165)

Evaluation:
...
....
--> STEP: 21
| > loss_disc: 2.4334442615509033 (2.429446754001436)
| > loss_disc_real_0: 0.12144245952367783 (0.13465816066378644)
| > loss_disc_real_1: 0.16504763066768646 (0.20793592716966355)
| > loss_disc_real_2: 0.2902933359146118 (0.2909027777966999)
| > loss_disc_real_3: 0.25102439522743225 (0.24391093992051624)
| > loss_disc_real_4: 0.3473176658153534 (0.24767487815448216)
| > loss_disc_real_5: 0.28220587968826294 (0.26010643371513914)
| > loss_0: 2.4334442615509033 (2.429446754001436)
| > loss_gen: 2.770388603210449 (2.59090789159139)
| > loss_kl: 1.5279167890548706 (2.3010178974696567)
| > loss_feat: 9.003446578979492 (7.517528613408406)
| > loss_mel: 25.329435348510742 (22.98664746965681)
| > loss_duration: 1.6732535362243652 (1.4985651345480056)
| > loss_1: 40.304439544677734 (36.894666853405184)

--> STEP: 22
| > loss_disc: 2.23518705368042 (2.420616767623208)
| > loss_disc_real_0: 0.07352343201637268 (0.13187930936163125)
| > loss_disc_real_1: 0.22937196493148804 (0.20891029252247376)
| > loss_disc_real_2: 0.3216579854488373 (0.29230074178088794)
| > loss_disc_real_3: 0.22647826373577118 (0.24311854554848236)
| > loss_disc_real_4: 0.3893234431743622 (0.25411344929174945)
| > loss_disc_real_3: 0.22647826373577118 (0.24311854554848236) [0/1848]
| > loss_disc_real_4: 0.3893234431743622 (0.25411344929174945)
| > loss_disc_real_5: 0.24885831773281097 (0.25959515571594244)
| > loss_0: 2.23518705368042 (2.420616767623208)
| > loss_gen: 3.2849197387695312 (2.622453884644942)
| > loss_kl: 1.5750868320465088 (2.2680210308595137)
| > loss_feat: 10.77985668182373 (7.6658162528818306)
| > loss_mel: 24.645959854125977 (23.062070759859953)
| > loss_duration: 1.6656222343444824 (1.506158639084209)
| > loss_1: 41.951446533203125 (37.12452047521418)

| > Synthesizing test sentences.
Input text cannot be None
! Run is kept in /home/dev/workspace/TTS/recipes/ljspeech/multi_speaker/speaker_multispeaker_fromscratch-January-17-2025_08+17PM-dbf1a08a
Traceback (most recent call last):
File "/home/dev/anaconda3/envs/tts2/lib/python3.9/site-packages/trainer/trainer.py", line 1833, in fit
self._fit()
File "/home/dev/anaconda3/envs/tts2/lib/python3.9/site-packages/trainer/trainer.py", line 1789, in _fit
self.test_run()
File "/home/dev/anaconda3/envs/tts2/lib/python3.9/site-packages/trainer/trainer.py", line 1698, in test_run
test_outputs = self.model.test_run(self.training_assets)
File "/home/dev/anaconda3/envs/tts2/lib/python3.9/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/home/dev/workspace/TTS/TTS/tts/models/vits.py", line 1442, in test_run
wav, alignment, _, _ = synthesis(
File "/home/dev/workspace/TTS/TTS/tts/utils/synthesis.py", line 221, in synthesis
outputs = run_model_torch(
File "/home/dev/workspace/TTS/TTS/tts/utils/synthesis.py", line 53, in run_model_torch
outputs = _func(
File "/home/dev/anaconda3/envs/tts2/lib/python3.9/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/home/dev/workspace/TTS/TTS/tts/models/vits.py", line 1150, in inference
attn = generate_path(w_ceil.squeeze(1), attn_mask.squeeze(1).transpose(1, 2))
IndexError: Dimension out of range (expected to be in range of [-2, 1], but got 2)`

To Reproduce

Run vits multispeaker for 14 spk

Expected behavior

No response

Logs

Environment

git+https://github.com/coqui-ai/TTS.git@dbf1a08a0d4e47fdad6172e433eeb34bc6b13b4e#egg=TTS

Additional context

No response

[eginhard]

Please only report an issue here if you're using the fork's code, the original package hasn't been updated in over a year now. Can you try again with the fork and provide enough details to reproduce (training recipe, config, environment).