Using replicate but it feels like my reference_file param isn't being trained on

[scoroman]

Hi, I've been using replicate to train my model on audio with 50 diffusion steps and it still sounds like the robotic default voice. Is my reference_file being recognized? Any help would be appreciated.

output = replicate.run(
model_id,
input={
"beta": 0.3,
"reference_file": "training_audio.WAV", # also tried: "training_audio.mp4" "training_audio.mp3"

"seed": 2,
"text": text_to_synthesize,
"alpha": 0.3,
"diffusion_steps": 50,
"embedding_scale": 5,

}
)
print(output)

[yl4579]

You may want to follow up with #97 or @alaradirik