feat: tts

:q

flake.nix

@@ -62,12 +62,25 @@
       system = "x86_64-linux";
     in
     {
-      packages.${system} = {
-        qwen-asr = nixpkgs.legacyPackages.${system}.callPackage ./packages/qwen-asr.nix { };
-        llama-swap = nixpkgs.legacyPackages.${system}.callPackage ./packages/llama-swap { };
-        z-image-models = nixpkgs.legacyPackages.${system}.callPackage ./packages/z-image-models { };
-        whisper-models = nixpkgs.legacyPackages.${system}.callPackage ./packages/whisper-models { };
-      };
+      packages.${system} =
+        let
+          lib = nixpkgs.lib;
+          pkgs = import nixpkgs {
+            inherit system;
+            config.allowUnfreePredicate =
+              pkg:
+              builtins.elem (lib.getName pkg) [
+                "fish-speech-models"
+              ];
+          };
+        in
+        {
+          qwen-asr = pkgs.callPackage ./packages/qwen-asr.nix { };
+          llama-swap = pkgs.callPackage ./packages/llama-swap { };
+          z-image-models = pkgs.callPackage ./packages/z-image-models { };
+          whisper-models = pkgs.callPackage ./packages/whisper-models { };
+          fish-speech-models = pkgs.callPackage ./packages/fish-speech-models { };
+        };
 
       # legolas
       nixosConfigurations.legolas = nixpkgs.lib.nixosSystem {

modules/llama-swap.nix

@@ -7,12 +7,14 @@
   ...
 }:
 {
+
   environment.systemPackages = [
     pkgs.unstable.llama-cpp-vulkan
     pkgs.unstable.stable-diffusion-cpp-vulkan
     pkgs.unstable.whisper-cpp-vulkan
     inputs.self.packages.${system}.z-image-models
     inputs.self.packages.${system}.whisper-models
+    inputs.self.packages.${system}.fish-speech-models
   ];
 
   services.llama-swap = {
@@ -28,6 +30,7 @@
         whisper-server = lib.getExe' pkgs.unstable.whisper-cpp-vulkan "whisper-server";
         z-image-models = inputs.self.packages.${system}.z-image-models;
         whisper-models = inputs.self.packages.${system}.whisper-models;
+        fish-speech-models = inputs.self.packages.${system}.fish-speech-models;
       in
       {
         healthCheckTimeout = 180;
@@ -125,6 +128,19 @@
             ];
             macros.model = "${whisper-models}/models/nb-whisper-small-q5_0.bin";
           };
+          "fish-speech-s2-pro" = {
+            cmd = "docker run --rm --name fish-speech-$\{PORT\} -p $\{PORT\}:8080 -v $\{models\}/s2-pro:/app/checkpoints/s2-pro:ro fishaudio/fish-speech:server-cpu";
+            cmdStop = "docker stop fish-speech-$\{PORT\}";
+            checkEndpoint = "/v1/health";
+            proxy = "http://127.0.0.1:$\{PORT\}";
+            ttl = 0;
+            aliases = [
+              "tts-1"
+              "tts"
+              "fish-speech"
+            ];
+            macros.models = "${fish-speech-models}/models";
+          };
         };
       };
   };

modules/python-packages.nix

@@ -6,6 +6,7 @@ ps: with ps; [
   numpy
   scipy
   pandas
+  polars
   matplotlib
   seaborn
   scikit-learn

packages/fish-speech-models/default.nix

@@ -0,0 +1,89 @@
+{
+  lib,
+  fetchurl,
+  runCommand,
+}:
+
+let
+  model1 = fetchurl {
+    url = "https://huggingface.co/fishaudio/s2-pro/resolve/main/model-00001-of-00002.safetensors";
+    hash = "sha256-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=";
+    name = "model-00001-of-00002.safetensors";
+  };
+
+  model2 = fetchurl {
+    url = "https://huggingface.co/fishaudio/s2-pro/resolve/main/model-00002-of-00002.safetensors";
+    hash = "sha256-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=";
+    name = "model-00002-of-00002.safetensors";
+  };
+
+  codec = fetchurl {
+    url = "https://huggingface.co/fishaudio/s2-pro/resolve/main/codec.pth";
+    hash = "sha256-dPxBxacVHG81Cvi9fl1uOsz8x/Pfv6wjr9Na8HBSuy8=";
+    name = "codec.pth";
+  };
+
+  config = fetchurl {
+    url = "https://huggingface.co/fishaudio/s2-pro/resolve/main/config.json";
+    hash = "sha256-JhtRmiqVdnEPyFM6dyl/rgB/Dns6ooohf4NSt/Mv6ZM=";
+    name = "config.json";
+  };
+
+  tokenizer = fetchurl {
+    url = "https://huggingface.co/fishaudio/s2-pro/resolve/main/tokenizer.json";
+    hash = "sha256-8k4ICZ1FqK3z9S9fCwMnbkM7udaJuxX8vMSM5YdEWIs=";
+    name = "tokenizer.json";
+  };
+
+  tokenizerConfig = fetchurl {
+    url = "https://huggingface.co/fishaudio/s2-pro/resolve/main/tokenizer_config.json";
+    hash = "sha256-uNFJNDrkJbDaZ+Zwhoas61G+eBXZeS8mX8Ev8E1emFY=";
+    name = "tokenizer_config.json";
+  };
+
+  specialTokensMap = fetchurl {
+    url = "https://huggingface.co/fishaudio/s2-pro/resolve/main/special_tokens_map.json";
+    hash = "sha256-wv8Y/ebkO3QIQ1vI7Qea90UxvvulSTWKl8vFnOYGvGs=";
+    name = "special_tokens_map.json";
+  };
+
+  chatTemplate = fetchurl {
+    url = "https://huggingface.co/fishaudio/s2-pro/resolve/main/chat_template.jinja";
+    hash = "sha256-h6JyjLjcn+Qk1iRUL2Bg7AWh0oXrvsV4uweJAOMzlrU=";
+    name = "chat_template.jinja";
+  };
+
+  modelIndex = fetchurl {
+    url = "https://huggingface.co/fishaudio/s2-pro/resolve/main/model.safetensors.index.json";
+    hash = "sha256-yMuZdNPRdmOpXbodbD6lMfxC03fA3/lG86AaDsSNRaM=";
+    name = "model.safetensors.index.json";
+  };
+in
+runCommand "fish-speech-models"
+  {
+    version = "1.0.0";
+    meta = {
+      description = "Fish Audio S2 Pro TTS model";
+      homepage = "https://huggingface.co/fishaudio/s2-pro";
+      license = {
+        shortName = "fish-audio-research";
+        fullName = "Fish Audio Research License";
+        url = "https://huggingface.co/fishaudio/s2-pro/blob/main/LICENSE.md";
+        free = false;
+        redistributable = true;
+      };
+      platforms = lib.platforms.all;
+    };
+  }
+  ''
+    mkdir -p $out/models/s2-pro
+    ln -s ${model1} $out/models/s2-pro/model-00001-of-00002.safetensors
+    ln -s ${model2} $out/models/s2-pro/model-00002-of-00002.safetensors
+    ln -s ${codec} $out/models/s2-pro/codec.pth
+    ln -s ${config} $out/models/s2-pro/config.json
+    ln -s ${tokenizer} $out/models/s2-pro/tokenizer.json
+    ln -s ${tokenizerConfig} $out/models/s2-pro/tokenizer_config.json
+    ln -s ${specialTokensMap} $out/models/s2-pro/special_tokens_map.json
+    ln -s ${chatTemplate} $out/models/s2-pro/chat_template.jinja
+    ln -s ${modelIndex} $out/models/s2-pro/model.safetensors.index.json
+  ''