dimension out of range (expected to be in range of [-2, 1], but got 2)

[S4lXLV]

I installed TTS through PyPi and I'm using this Python code to do the job for me:

# Import the TTS API
from TTS.api import TTS

# Initialize TTS with the target model name and vocoder path
tts = TTS(model_name="tts_models/en/ljspeech/vits--neon",
          vocoder_path="vocoder_models/en/ljspeech/hifigan_v2",
          progress_bar=True,
          gpu=True)

# Read input text from file
with open('File', 'r', encoding='utf-8') as file:
    input_text = file.read()

# Remove all Unicode characters from the input text
input_text = input_text.encode("ascii", "ignore").decode()

# Generate speech for the input text and save it to a WAV file
tts.tts_to_file(text=input_text, file_path='Audio.wav')

it kinda works as it sometimes gets the job done but usually, I get hit with a dimension out of range (expected to be in range of [-2, 1], but got 2)

As you can see i added input_text.encode("ascii", "ignore").decode() And encoding='utf-8 so it might solve the problem but nothing is working. What am i missing here?

Full logs:

IndexError                                Traceback (most recent call last)
Cell In [1], line 19
     16 input_text = input_text.encode("ascii", "ignore").decode()
     18 # Generate speech for the input text and save it to a WAV file
---> 19 tts.tts_to_file(text=input_text, file_path='/notebooks/part2.wav')

File /usr/local/lib/python3.9/dist-packages/TTS/api.py:573, in TTS.tts_to_file(self, text, speaker, language, speaker_wav, emotion, speed, file_path)
    569 if self.csapi is not None:
    570     return self.tts_coqui_studio(
    571         text=text, speaker_name=speaker, language=language, emotion=emotion, speed=speed, file_path=file_path
    572     )
--> 573 wav = self.tts(text=text, speaker=speaker, language=language, speaker_wav=speaker_wav)
    574 self.synthesizer.save_wav(wav=wav, path=file_path)
    575 return file_path

File /usr/local/lib/python3.9/dist-packages/TTS/api.py:524, in TTS.tts(self, text, speaker, language, speaker_wav, emotion, speed)
    519 if self.csapi is not None:
    520     return self.tts_coqui_studio(
    521         text=text, speaker_name=speaker, language=language, emotion=emotion, speed=speed
    522     )
--> 524 wav = self.synthesizer.tts(
    525     text=text,
    526     speaker_name=speaker,
    527     language_name=language,
    528     speaker_wav=speaker_wav,
    529     reference_wav=None,
    530     style_wav=None,
    531     style_text=None,
    532     reference_speaker_name=None,
    533 )
    534 return wav

File /usr/local/lib/python3.9/dist-packages/TTS/utils/synthesizer.py:323, in Synthesizer.tts(self, text, speaker_name, language_name, speaker_wav, style_wav, style_text, reference_wav, reference_speaker_name)
    320 if not reference_wav:
    321     for sen in sens:
    322         # synthesize voice
--> 323         outputs = synthesis(
    324             model=self.tts_model,
    325             text=sen,
    326             CONFIG=self.tts_config,
    327             use_cuda=self.use_cuda,
    328             speaker_id=speaker_id,
    329             style_wav=style_wav,
    330             style_text=style_text,
    331             use_griffin_lim=use_gl,
    332             d_vector=speaker_embedding,
    333             language_id=language_id,
    334         )
    335         waveform = outputs["wav"]
    336         mel_postnet_spec = outputs["outputs"]["model_outputs"][0].detach().cpu().numpy()

File /usr/local/lib/python3.9/dist-packages/TTS/tts/utils/synthesis.py:213, in synthesis(model, text, CONFIG, use_cuda, speaker_id, style_wav, style_text, use_griffin_lim, do_trim_silence, d_vector, language_id)
    211 text_inputs = text_inputs.unsqueeze(0)
    212 # synthesize voice
--> 213 outputs = run_model_torch(
    214     model,
    215     text_inputs,
    216     speaker_id,
    217     style_mel,
    218     style_text,
    219     d_vector=d_vector,
    220     language_id=language_id,
    221 )
    222 model_outputs = outputs["model_outputs"]
    223 model_outputs = model_outputs[0].data.cpu().numpy()

File /usr/local/lib/python3.9/dist-packages/TTS/tts/utils/synthesis.py:50, in run_model_torch(model, inputs, speaker_id, style_mel, style_text, d_vector, language_id)
     48 else:
     49     _func = model.inference
---> 50 outputs = _func(
     51     inputs,
     52     aux_input={
     53         "x_lengths": input_lengths,
     54         "speaker_ids": speaker_id,
     55         "d_vectors": d_vector,
     56         "style_mel": style_mel,
     57         "style_text": style_text,
     58         "language_ids": language_id,
     59     },
     60 )
     61 return outputs

File /usr/local/lib/python3.9/dist-packages/torch/utils/_contextlib.py:115, in context_decorator.<locals>.decorate_context(*args, **kwargs)
    112 @functools.wraps(func)
    113 def decorate_context(*args, **kwargs):
    114     with ctx_factory():
--> 115         return func(*args, **kwargs)

File /usr/local/lib/python3.9/dist-packages/TTS/tts/models/vits.py:1149, in Vits.inference(self, x, aux_input)
   1146 y_mask = sequence_mask(y_lengths, None).to(x_mask.dtype).unsqueeze(1)  # [B, 1, T_dec]
   1148 attn_mask = x_mask * y_mask.transpose(1, 2)  # [B, 1, T_enc] * [B, T_dec, 1]
-> 1149 attn = generate_path(w_ceil.squeeze(1), attn_mask.squeeze(1).transpose(1, 2))
   1151 m_p = torch.matmul(attn.transpose(1, 2), m_p.transpose(1, 2)).transpose(1, 2)
   1152 logs_p = torch.matmul(attn.transpose(1, 2), logs_p.transpose(1, 2)).transpose(1, 2)

IndexError: Dimension out of range (expected to be in range of [-2, 1], but got 2)

[knochenhans]

Hi, I’ve been running into the exact same Exception raised by attn = generate_path(w_ceil.squeeze(1), attn_mask.squeeze(1).transpose(1, 2)) lately. I haven’t been able to reproduce this bug reliably, trying to synthesize the same text a second time usually works.

I have a feeling this came up with one of the latest updates of Coqui.ai TTS, as I’ve been working with this for about a year now and this specific bug problem came up first a few weeks ago for me on both of my machines. Maybe downgrading to a pre-0.13 release helps?

[Hexatona]

I know what happens here. The short answer is, when chopping up text into sentences, sometimes it grabs something like a quote or something as it's own sentence - it tries to synthesize it and runs into an error.

My solution suggestions: catch the error, and either try the same text again, or just try that bit again after removing superfluous text.

[knochenhans]

For some reason it never occurred to me I could just catch the error and repeat the synthesization in my library, so thanks for the idea 😅

[cod3r0k]

I, Did you solve this issue?
I have the same as you said.

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)