https://www.dafei.me/tags/omnivoice/ Recent content in OmniVoice on 大飞的博客 Hugo zh-cn Mon, 25 May 2026 00:00:00 +0000 https://www.dafei.me/posts/voice-01-omnivoice/ Mon, 25 May 2026 00:00:00 +0000 https://www.dafei.me/posts/voice-01-omnivoice/ <p>想让 AI 用你的声音说话？只需要一段 3 到 10 秒的录音，OmniVoice 就能克隆你的声音，说出任意文字。</p> <hr> <h2 id="omnivoice-是什么">OmniVoice 是什么</h2> <p>OmniVoice 是小米 / k2-fsa 团队开发的开源零样本 TTS 模型，支持 600+ 种语言，基于<strong>扩散语言模型</strong>架构，推理速度极快（RTF 最低 0.025，比实时快 40 倍）。</p> <p>支持三种模式：</p> <ul> <li><strong>语音克隆</strong>：提供参考音频，克隆声音说出任意文字</li> <li><strong>语音设计</strong>：用文字描述声音（性别、年龄、音调、口音等）</li> <li><strong>自动语音</strong>：模型自动选择声音</li> </ul> <hr> <h2 id="环境准备">环境准备</h2> <h3 id="安装-conda-环境">安装 conda 环境</h3> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"><code class="language-bash" data-lang="bash"><span style="display:flex;"><span>conda create -n omnivoice python<span style="color:#f92672">=</span>3.11 -y </span></span><span style="display:flex;"><span>conda activate omnivoice </span></span></code></pre></div><h3 id="安装-pytorch">安装 PyTorch</h3> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"><code class="language-bash" data-lang="bash"><span style="display:flex;"><span><span style="color:#75715e"># Apple Silicon</span> </span></span><span style="display:flex;"><span>pip install torch<span style="color:#f92672">==</span>2.8.0 torchaudio<span style="color:#f92672">==</span>2.8.0 </span></span><span style="display:flex;"><span> </span></span><span style="display:flex;"><span><span style="color:#75715e"># NVIDIA GPU（以 CUDA 12.8 为例）</span> </span></span><span style="display:flex;"><span>pip install torch<span style="color:#f92672">==</span>2.8.0+cu128 torchaudio<span style="color:#f92672">==</span>2.8.0+cu128 --extra-index-url https://download.pytorch.org/whl/cu128 </span></span></code></pre></div><h3 id="安装-omnivoice">安装 OmniVoice</h3> <div class="highlight"><pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"><code class="language-bash" data-lang="bash"><span style="display:flex;"><span>git clone https://github.com/k2-fsa/OmniVoice.git </span></span><span style="display:flex;"><span>cd OmniVoice </span></span><span style="display:flex;"><span>pip install -e . </span></span></code></pre></div><hr> <h2 id="下载模型">下载模型</h2> <p>第一次运行会自动下载。如果在国内连不上 HuggingFace，设置镜像：</p>