summary refs log tree commit diff
path: root/pkgs/tools/audio/kaldi
diff options
context:
space:
mode:
authorJörg Thalheim <joerg@thalheim.io>2020-12-31 09:24:02 +0100
committerJörg Thalheim <joerg@thalheim.io>2021-01-01 11:07:37 +0100
commite03e3c5fad8b56d1f3f252ea92271dadc639f4aa (patch)
tree803fbf2c81db0b1a5a83b2245b1f7a5e1b61b33d /pkgs/tools/audio/kaldi
parent68a8e7bb26d62125b6551bc10e198eef8f057e34 (diff)
downloadnixpkgs-e03e3c5fad8b56d1f3f252ea92271dadc639f4aa.tar
nixpkgs-e03e3c5fad8b56d1f3f252ea92271dadc639f4aa.tar.gz
nixpkgs-e03e3c5fad8b56d1f3f252ea92271dadc639f4aa.tar.bz2
nixpkgs-e03e3c5fad8b56d1f3f252ea92271dadc639f4aa.tar.lz
nixpkgs-e03e3c5fad8b56d1f3f252ea92271dadc639f4aa.tar.xz
nixpkgs-e03e3c5fad8b56d1f3f252ea92271dadc639f4aa.tar.zst
nixpkgs-e03e3c5fad8b56d1f3f252ea92271dadc639f4aa.zip
kaldi: init at 2020-12-26
State-of-the-art speech-to-text engine - used in rhasspy

Co-authored-by: Sandro <sandro.jaeckel@gmail.com>
Diffstat (limited to 'pkgs/tools/audio/kaldi')
-rw-r--r--pkgs/tools/audio/kaldi/default.nix88
1 files changed, 88 insertions, 0 deletions
diff --git a/pkgs/tools/audio/kaldi/default.nix b/pkgs/tools/audio/kaldi/default.nix
new file mode 100644
index 00000000000..fa291a05078
--- /dev/null
+++ b/pkgs/tools/audio/kaldi/default.nix
@@ -0,0 +1,88 @@
+{ stdenv
+, openblas
+, blas
+, lapack
+, openfst
+, icu
+, cmake
+, pkg-config
+, fetchFromGitHub
+, git
+, python3
+}:
+
+assert blas.implementation == "openblas" && lapack.implementation == "openblas";
+let
+  # rev from https://github.com/kaldi-asr/kaldi/blob/master/cmake/third_party/openfst.cmake
+  openfst = fetchFromGitHub {
+    owner = "kkm000";
+    repo = "openfst";
+    rev = "0bca6e76d24647427356dc242b0adbf3b5f1a8d9";
+    sha256 = "1802rr14a03zl1wa5a0x1fa412kcvbgprgkadfj5s6s3agnn11rx";
+  };
+in
+stdenv.mkDerivation {
+  pname = "kaldi";
+  version = "2020-12-26";
+
+  src = fetchFromGitHub {
+    owner = "kaldi-asr";
+    repo = "kaldi";
+    rev = "813b73185a18725e4f6021981d17221d6ee23a19";
+    sha256 = "sha256-lTqXTG5ZTPmhCgt+BVzOwjKEIj+bLGUa+IxJq+XtHUg=";
+  };
+
+  cmakeFlags = [
+    "-DKALDI_BUILD_TEST=off"
+    "-DBUILD_SHARED_LIBS=on"
+  ];
+
+  preConfigure = ''
+    mkdir bin
+    cat > bin/git <<'EOF'
+    #!${stdenv.shell}
+    if [[ "$1" == "--version" ]]; then
+      # cmake checks this
+      ${git}/bin/git --version
+    elif [[ "$1" == "clone" ]]; then
+      # mock this call:
+
+      # https://github.com/kaldi-asr/kaldi/blob/c9d8b9ad3fef89237ba5517617d977b7d70a7ed5/cmake/third_party/openfst.cmake#L5
+      cp -r ${openfst} ''${@: -1}
+      chmod -R +w ''${@: -1}
+    elif [[ "$1" == "rev-list" ]]; then
+      # fix up this call:
+      # https://github.com/kaldi-asr/kaldi/blob/c9d8b9ad3fef89237ba5517617d977b7d70a7ed5/cmake/VersionHelper.cmake#L8
+      echo 0
+    fi
+    true
+    EOF
+    chmod +x bin/git
+    export PATH=$(pwd)/bin:$PATH
+  '';
+
+  buildInputs = [
+    openblas
+    openfst
+    icu
+  ];
+
+  nativeBuildInputs = [
+    cmake
+    pkg-config
+    python3
+  ];
+
+  postInstall = ''
+    mkdir -p $out/share/kaldi
+    cp -r ../egs $out/share/kaldi
+  '';
+
+  meta = with stdenv.lib; {
+    description = "Speech Recognition Toolkit";
+    homepage = "https://kaldi-asr.org";
+    license = licenses.mit;
+    maintainers = with maintainers; [ mic92 ];
+    platforms = platforms.linux;
+  };
+}