base_model: unsloth/orpheus-3b-0.1-ft-unsloth-bnb-4bit tags:

• transformers
• unsloth
• llama
• trl
• tts
• tex-to-speech license: apache-2.0 language:
• pl datasets:
• czyzi0/the-mc-speech-dataset pipeline_tag: text-to-speech

VoxPolska：新一代波兰语语音生成

📌 模型亮点

• 语境感知语音：生成能捕捉波兰语细微差别和语调的语音。
• 展现语音合成和波兰语处理的高级能力。
• 将波兰语书面文本转换为自然、流畅且富有表现力的语音。
• 先进深度学习：采用尖端深度学习技术实现最佳性能。
• 前沿技术：展示语音合成和波兰语处理的高级能力。

⚙️ 技术细节

• 基础模型：Orpheus TTS
• 应用LoRA（低秩适配）微调以优化模型性能。
• 采样率：24 kHz音频输出，实现高保真音质。
• 训练数据：24000+波兰语文本-音频对
• 合并16位量化
• 音频解码：定制分层处理以生成音频
• 重复惩罚：1.1，避免重复短语
• 梯度检查点：启用以高效利用内存

🎧 使用示例

• 以下是在笔记本上运行模型的示例代码

  !pip install snac torch transformers

  import torch
  import snac
  from snac import SNAC
  from transformers import AutoTokenizer, AutoModelForCausalLM
  import os
  from IPython.display import display, Audio
  
  device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
  
  tokenizer = AutoTokenizer.from_pretrained("salihfurkaan/VoxPolska-V1-Merged-16bit")
  model = AutoModelForCausalLM.from_pretrained("salihfurkaan/VoxPolska-V1-Merged-16bit").to(device)
  
  os.environ["HF_TOKEN"] = "在此输入您的huggingface令牌"
  snac_model = SNAC.from_pretrained("hubertsiuzdak/snac_24khz").to(device)
  
  prompts = [
      "Cześć, jestem dużym modelem języka sztucznej inteligencji"
  ]  #示例提示
  chosen_voice = None
  
  prompts_ = [(f"{chosen_voice}: " + p) if chosen_voice else p for p in prompts]
  all_input_ids = []
  for prompt in prompts_:
      input_ids = tokenizer(prompt, return_tensors="pt").input_ids
      all_input_ids.append(input_ids)
  
  start_token = torch.tensor([[128259]], dtype=torch.int64)  # 人类开始标记
  end_tokens = torch.tensor([[128009, 128260]], dtype=torch.int64)  # 文本结束标记, 人类结束标记
  
  all_modified_input_ids = []
  for input_ids in all_input_ids:
      modified_input_ids = torch.cat([start_token, input_ids, end_tokens], dim=1)
      all_modified_input_ids.append(modified_input_ids)
  
  all_padded_tensors = []
  all_attention_masks = []
  max_length = max([x.shape[1] for x in all_modified_input_ids])
  for modified_input_ids in all_modified_input_ids:
      padding = max_length - modified_input_ids.shape[1]
      padded_tensor = torch.cat([torch.full((1, padding), 128263, dtype=torch.int64), modified_input_ids], dim=1)
      attention_mask = torch.cat([torch.zeros((1, padding), dtype=torch.int64), torch.ones((1, modified_input_ids.shape[1]), dtype=torch.int64)], dim=1)
      all_padded_tensors.append(padded_tensor)
      all_attention_masks.append(attention_mask)
  
  all_padded_tensors = torch.cat(all_padded_tensors, dim=0).to(device)
  all_attention_masks = torch.cat(all_attention_masks, dim=0).to(device)
  
  generated_ids = model.generate(
      input_ids=all_padded_tensors,
      attention_mask=all_attention_masks,
      max_new_tokens=1200,
      do_sample=True,
      temperature=0.6,
      top_p=0.95,
      repetition_penalty=1.1,
      num_return_sequences=1,
      eos_token_id=128258,
      use_cache=True
  )
  
  token_to_find = 128257
  token_to_remove = 128258
  token_indices = (generated_ids == token_to_find).nonzero(as_tuple=True)
  
  if len(token_indices[1]) > 0:
      last_occurrence_idx = token_indices[1][-1].item()
      cropped_tensor = generated_ids[:, last_occurrence_idx+1:]
  else:
      cropped_tensor = generated_ids
  
  processed_rows = []
  for row in cropped_tensor:
      masked_row = row[row != token_to_remove]
      processed_rows.append(masked_row)
  
  code_lists = []
  for row in processed_rows:
      row_length = row.size(0)
      new_length = (row_length // 7) * 7
      trimmed_row = row[:new_length]
      trimmed_row = [t - 128266 for t in trimmed_row]
      code_lists.append(trimmed_row)
  
  def redistribute_codes(code_list):
      layer_1 = []
      layer_2 = []
      layer_3 = []
      for i in range((len(code_list) + 1) // 7):
          layer_1.append(code_list[7 * i])
          layer_2.append(code_list[7 * i + 1] - 4096)
          layer_3.append(code_list[7 * i + 2] - (2 * 4096))
          layer_3.append(code_list[7 * i + 3] - (3 * 4096))
          layer_2.append(code_list[7 * i + 4] - (4 * 4096))
          layer_3.append(code_list[7 * i + 5] - (5 * 4096))
          layer_3.append(code_list[7 * i + 6] - (6 * 4096))
      
      codes = [
          torch.tensor(layer_1).unsqueeze(0).to(device),
          torch.tensor(layer_2).unsqueeze(0).to(device),
          torch.tensor(layer_3).unsqueeze(0).to(device)
      ]
      audio_hat = snac_model.decode(codes)
      return audio_hat
      
  my_samples = []
  for code_list in code_lists:
      samples = redistribute_codes(code_list)
      my_samples.append(samples)
  
  if len(prompts) != len(my_samples):
      raise Exception("提示数量与样本数量不匹配")
  else:
      for i in range(len(my_samples)):
          print(prompts[i])
          samples = my_samples[i]
          display(Audio(samples.detach().squeeze().to("cpu").numpy(), rate=24000))
  
  del my_samples, samples

您可以从此处获取您的huggingface令牌

📬 联系与支持

如有问题、建议和反馈，请在HuggingFace上提交问题。您也可以通过以下方式联系作者： LinkedIn

模型滥用

请勿在未经同意的情况下使用此模型进行冒充、传播错误信息或欺骗（包括假新闻或欺诈性电话），或任何非法或有害活动。使用此模型即表示您同意遵守所有适用的法律和道德准则。

引用

@misc{
  title={salihfurkaan/VoxPolska-V1-Merged-16bit},
  author={Salih Furkan Erik},
  year={2025},
  url={https://huggingface.co/salihfurkaan/VoxPolska-V1-Merged-16bit/}
}