[Bug] tts_with_vc_to_file uses cuda even though "cpu" is specified as device

Describe the bug

The model tries to use CUDA, even though I specified it to use the CPU. Because I have one of those computers with a soldered on GPU with only 4GB of memory, that is bound to crash and burn as soon as it tries to allocate some memory.

To Reproduce

from TTS.api import TTS

tts = TTS(model_name="tts_models/de/thorsten/tacotron2-DDC", progress_bar=False).to("cpu") tts.tts_with_vc_to_file( speaker_wav="my_speaker.wav", text="Hallo! Das ist ein Test", file_path="output.wav")

Expected behavior

The model should run on the cpu.

Logs

--truncated--

 > voice_conversion_models/multilingual/vctk/freevc24 is already downloaded.
 > Using model: freevc
 > Loading pretrained speaker encoder model ...
Loaded the voice encoder model on cuda in 0.19 seconds.
Traceback (most recent call last):
  File "my_script.py", line 20, in <module>
    tts.tts_with_vc_to_file(
  File ".venv/lib/python3.11/site-packages/TTS/api.py", line 455, in tts_with_vc_to_file
    wav = self.tts_with_vc(
          ^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/TTS/api.py", line 420, in tts_with_vc
    wav = self.voice_converter.voice_conversion(source_wav=fp.name, target_wav=speaker_wav)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/TTS/utils/synthesizer.py", line 254, in voice_conversion
    output_wav = self.vc_model.voice_conversion(source_wav, target_wav)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/TTS/vc/models/freevc.py", line 523, in voice_conversion
    g_tgt = self.enc_spk_ex.embed_utterance(wav_tgt)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/TTS/vc/modules/freevc/speaker_encoder/speaker_encoder.py", line 155, in embed_utterance
    partial_embeds = self(mels).cpu().numpy()
                     ^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/TTS/vc/modules/freevc/speaker_encoder/speaker_encoder.py", line 60, in forward
    _, (hidden, _) = self.lstm(mels)
                     ^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/torch/nn/modules/rnn.py", line 911, in forward
    result = _VF.lstm(input, hx, self._flat_weights, self.bias, self.num_layers,
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 7.35 GiB. GPU

Environment

{
    "CUDA": {
        "GPU": [
            "NVIDIA RTX A500 Laptop GPU"
        ],
        "available": true,
        "version": "12.1"
    },
    "Packages": {
        "PyTorch_debug": false,
        "PyTorch_version": "2.3.1+cu121",
        "TTS": "0.22.0",
        "numpy": "1.26.4"
    },
    "System": {
        "OS": "Linux",
        "architecture": [
            "64bit",
            "ELF"
        ],
        "processor": "x86_64",
        "python": "3.11.9",
        "version": "#35-Ubuntu SMP PREEMPT_DYNAMIC Mon May 20 15:51:52 UTC 2024"
    }
}

Additional context

No response

coqui-ai / TTS