Stacking two tensors from different prackages Kokora and OuteTTS

gistfile1.txt

import outetts
import wave
import os
import torch
import numpy as np

# Configure the model
model_config = outetts.HFModelConfig_v1(model_path="OuteAI/OuteTTS-0.2-500M", language="en")
interface = outetts.InterfaceHF(model_version="0.2", cfg=model_config)
interface.print_default_speakers()

def create_speaker(sample="output.m4a"):
    speaker = interface.create_speaker(audio_path=sample, transcript=None, whisper_model="turbo", whisper_device=None)
    return speaker

def load_speaker(speaker_file):
    embedding = torch.load(speaker_file)
    return interface.create_speaker_from_embedding(embedding)

def save_tensor(speaker_dict, output_file):
    codes_list = [word['codes'] for word in speaker_dict['words']]
    max_len = max(len(codes) for codes in codes_list)
    padded_codes = [codes + [0] * (max_len - len(codes)) for codes in codes_list]
    codes = np.array(padded_codes)   
    codes = codes.astype(np.float32) / 2000.0 - 0.5
    codes_reshaped = np.zeros((511, 1, 256), dtype=np.float32)
    codes_reshaped[:min(511, codes.shape[0]), 0, :min(256, codes.shape[1])] = codes[:511, :256]
    codes_tensor = torch.from_numpy(codes_reshaped).to(emma.dtype)
    torch.save(codes_tensor, output_file)

speaker_file = "bf_hannah.txt"
if not os.path.exists(speaker_file):
    speaker = create_speaker()
    interface.save_speaker(speaker, speaker_file)

speaker = interface.load_speaker(speaker_file)
emma = torch.load('bf_emma.pt', weights_only=True)

save_tensor(speaker, 'bf_hannah.pt')

hannah = torch.load('bf_hannah.pt', weights_only=True)

print(emma.dtype, hannah.dtype)
print(type(emma), type(hannah))

averaged_tensor = torch.mean(torch.stack([emma, hannah]), dim=0)
torch.save(averaged_tensor, "hammah.pth")