'GPT2InferenceModel' object has no attribute 'generate'

[jainsakku]

Hey All, I am trying to use TTS package, when calling the function wav = tts.tts(text="Hello world!", speaker_wav="test.mp3", language="en") I am getting this error. I am new to this stuff so please help me in resolving the issue.

[eginhard]

This is fixed in our fork, available via pip install coqui-tts. This repo is not maintained anymore.

[MohamedAziz15]

Yes, I'm looking in TTS Model to solve

from IPython.display import Audio, display
text = "صباح الخير"
print("Inference...")
out = model.forward(
   text,
   "ar",
   gpt_cond_latent,
   speaker_embedding,
   temperature=0.75,
)

AUDIO_OUTPUT_PATH = "/kaggle/working/EGTTS-V0.1/output_audio.wav"
torchaudio.save("xtts_audio.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)
display(Audio(AUDIO_OUTPUT_PATH, autoplay=True))

error:'GPT2InferenceModel' object has no attribute 'generate'

[eginhard]

You're likely still not using the fork or its latest version. Check you're on coqui-tts version 0.27.0

[MohamedAziz15]

yup, It's old version 0.22.0
What do you recommend me to do ?

[eginhard]

https://coqui-tts.readthedocs.io/en/latest/faq.html#what-is-the-coqui-fork-about-and-how-to-install-it

[MohamedAziz15]

thanks @eginhard , it's worked using :

git clone https://github.com/idiap/coqui-ai-TTS
cd coqui-ai-TTS
pip install -e .