babysor/MockingBird · error · Exception

Please load hifi-gan in memory before using it

Error message

Please load hifi-gan in memory before using it

What it means

infer_waveform in models/vocoder/hifigan/inference.py requires the module-global `generator` to be initialized by a prior load_model() call; otherwise it raises immediately.

Source

Thrown at models/vocoder/hifigan/inference.py:63

        _device = torch.device('cpu')

    generator = Generator(h).to(_device)
    state_dict_g = load_checkpoint(
        weights_fpath, _device
    )
    generator.load_state_dict(state_dict_g['generator'])
    generator.eval()
    generator.remove_weight_norm()


def is_loaded():
    return generator is not None


def infer_waveform(mel, progress_callback=None):

    if generator is None:
        raise Exception("Please load hifi-gan in memory before using it")

    mel = torch.FloatTensor(mel).to(_device)
    mel = mel.unsqueeze(0)

    with torch.no_grad():
        y_g_hat = generator(mel)
        audio = y_g_hat.squeeze()
    audio = audio.cpu().numpy()

    return audio, output_sample_rate

View on GitHub (pinned to 28dc5e14f1)

Solutions

1. Call hifigan.load_model(weights_fpath) before infer_waveform
2. Confirm the .pt/.pth checkpoint path is correct and the file loads
3. Guard with is_loaded() before calling

Example fix

# before
wav = hifigan.infer_waveform(mel)

# after
higigan = hifigan.load_model(Path('logs/hifigan/generator_v1'))
wav = hifigan.infer_waveform(mel)

from models.vocoder import hifigan
if not hifigan.is_loaded():
    hifigan.load_model(Path('models/vocoder/hifigan/generator_v1'))

def hifigan_ready() -> bool:
    from models.vocoder.hifigan import inference
    return inference.generator is not None

try:
    wav = hifigan.infer_waveform(mel)
except Exception as e:
    if 'load hifi-gan' in str(e):
        hifigan.load_model(ckpt); wav = hifigan.infer_waveform(mel)
    else: raise

When it happens

Trigger: Calling infer_waveform(mel) without a preceding successful load_model(...) in the same process.

Common situations: Skipping vocoder init in a custom synthesis script; failed/aborted checkpoint load; toolbox configured with missing hifgan weights.

Related errors

• Please load fre-gan in memory before using it
• Please load Wave-RNN in memory before using it
• Package 'webrtcvad' not found. This package enables noise re
• Model folder {VOC_MODELS_DIRT} doesn't exist.
• Model folder {VOC_MODELS_DIRT} doesn't exist.

AI-assisted analysis of babysor/MockingBird@28dc5e14f1 (2026-08-27). Data as JSON: /api/errors/cfc778b36a963015. Report an issue: GitHub.