in f5-tts_finetune-gradio how correct IndexError: too many indices for tensor of dimension 2

[paulokanda]

Hi,
any Idea about this error in test model?

in f5-tts_finetune-gradio how correct IndexError: too many indices for tensor of dimension 2

thanks in advance

[ephr0n]

Having the same issue, did you ever find a fix?

[Valoria191]

same issue, if somone know how to fix it please

[Valoria191]

I couldn't solve the problem, but I can suggest an alternative solution with python:
(found with chatGPT, work for me)

import torch
from tortoise.api import TextToSpeech
import torchaudio

model_path = "C:/Users/......../....../F5-TTS/ckpts/your_model/model_last.pt"

tts = TextToSpeech()

custom_state = torch.load(model_path, map_location="cpu")

tts.autoregressive.load_state_dict(custom_state, strict=False)

text = "Hello, I'm a personalised text-to-speech model."

audio = tts.tts_with_preset(
text,
preset="fast", # Choice : "fast", "standard", "high_quality"
)

audio = torch.tensor(audio).squeeze().unsqueeze(0) # Supprime les dimensions inutiles et garde le bon format

output_path = "C:/Users/casca/......../....../audio/audio.wav"
sample_rate = 24000
torchaudio.save(output_path, audio, sample_rate)

print(f"Voice synthesis saved in : {output_path}")