3 files changed, 141 insertions, 0 deletions
diff --git a/nixpkgs/pkgs/tools/audio/piper/default.nix b/nixpkgs/pkgs/tools/audio/piper/default.nix
new file mode 100644
index 000000000000..d753d7bb3e1e
--- /dev/null
+++ b/nixpkgs/pkgs/tools/audio/piper/default.nix
@@ -0,0 +1,69 @@
+{ lib
+, stdenv
+, fetchFromGitHub
+
+# build time
+, cmake
+, pkg-config
+
+# runtime
+, onnxruntime
+, pcaudiolib
+, piper-phonemize
+, spdlog
+
+# tests
+, piper-train
+}:
+
+stdenv.mkDerivation (finalAttrs: {
+  pname = "piper";
+  version = "1.2.0";
+
+  src = fetchFromGitHub {
+    owner = "rhasspy";
+    repo = "piper";
+    rev = "refs/tags/v${finalAttrs.version}";
+    hash = "sha256-6WNWqJt0PO86vnf+3iHaRRg2KwBOEj4aicmB+P2phlk=";
+  };
+
+  sourceRoot = "${finalAttrs.src.name}/src/cpp";
+
+  nativeBuildInputs = [
+    cmake
+    pkg-config
+  ];
+
+  buildInputs = [
+    onnxruntime
+    pcaudiolib
+    piper-phonemize
+    piper-phonemize.espeak-ng
+    spdlog
+  ];
+
+  env.NIX_CFLAGS_COMPILE = builtins.toString [
+    "-isystem ${lib.getDev piper-phonemize}/include/piper-phonemize"
+  ];
+
+  installPhase = ''
+    runHook preInstall
+
+    mkdir -p $out/bin
+    install -m 0755 piper $out/bin
+
+    runHook postInstall
+  '';
+
+  passthru.tests = {
+    inherit piper-train;
+  };
+
+  meta = with lib; {
+    changelog = "https://github.com/rhasspy/piper/releases/tag/v${finalAttrs.version}";
+    description = "A fast, local neural text to speech system";
+    homepage = "https://github.com/rhasspy/piper";
+    license = licenses.mit;
+    maintainers = with maintainers; [ hexa ];
+  };
+})
diff --git a/nixpkgs/pkgs/tools/audio/piper/fix-compilation-with-newer-onnxruntime.patch b/nixpkgs/pkgs/tools/audio/piper/fix-compilation-with-newer-onnxruntime.patch
new file mode 100644
index 000000000000..9d2e46bb4e91
--- /dev/null
+++ b/nixpkgs/pkgs/tools/audio/piper/fix-compilation-with-newer-onnxruntime.patch
@@ -0,0 +1,18 @@
+diff --git a/src/cpp/synthesize.hpp b/src/cpp/synthesize.hpp
+index ef61aef..4c7db7a 100644
+--- a/synthesize.hpp
++++ b/synthesize.hpp
+@@ -119,11 +119,11 @@ void synthesize(SynthesisConfig &synthesisConfig, ModelSession &session,
+ 
+   // Clean up
+   for (size_t i = 0; i < outputTensors.size(); i++) {
+-    Ort::OrtRelease(outputTensors[i].release());
++    Ort::detail::OrtRelease(outputTensors[i].release());
+   }
+ 
+   for (size_t i = 0; i < inputTensors.size(); i++) {
+-    Ort::OrtRelease(inputTensors[i].release());
++    Ort::detail::OrtRelease(inputTensors[i].release());
+   }
+ }
+ } // namespace larynx
diff --git a/nixpkgs/pkgs/tools/audio/piper/train.nix b/nixpkgs/pkgs/tools/audio/piper/train.nix
new file mode 100644
index 000000000000..2cab1ba4ba04
--- /dev/null
+++ b/nixpkgs/pkgs/tools/audio/piper/train.nix
@@ -0,0 +1,54 @@
+{ piper-tts
+, python3
+}:
+
+let
+  python = python3.override {
+    packageOverrides = self: super: {
+    };
+  };
+in
+
+python.pkgs.buildPythonPackage {
+  inherit (piper-tts) version src;
+
+  pname = "piper-train";
+  format = "setuptools";
+
+  sourceRoot = "${piper-tts.src.name}/src/python";
+
+  nativeBuildInputs = with python.pkgs; [
+    cython
+  ];
+
+  postBuild = ''
+    make -C piper_train/vits/monotonic_align
+  '';
+
+  postInstall = ''
+    export MONOTONIC_ALIGN=$out/${python.sitePackages}/piper_train/vits/monotonic_align/monotonic_align
+    mkdir -p $MONOTONIC_ALIGN
+    cp -v ./piper_train/vits/monotonic_align/piper_train/vits/monotonic_align/core.*.so $MONOTONIC_ALIGN/
+  '';
+
+  propagatedBuildInputs = with python.pkgs; [
+    espeak-phonemizer
+    librosa
+    numpy
+    onnxruntime
+    piper-phonemize
+    pytorch-lightning
+    torch
+  ];
+
+  pythonImportsCheck = [
+    "piper_train"
+  ];
+
+  doCheck = false; # no tests
+
+  meta = piper-tts.meta // {
+    # requires torch<2, pytorch-lightning~=1.7
+    broken = true;
+  };
+}