nixpkgs/pkgs/tools/audio/tts/default.nix

112 lines
3.1 KiB
Nix
Raw Normal View History

2020-06-08 10:38:55 -04:00
{ lib
, python3Packages
, fetchFromGitHub
, python3
}:
2021-01-16 08:11:12 -05:00
# USAGE:
# $ tts-server --list_models
# # pick your favorite vocoder/tts model
# $ tts-server --model_name tts_models/en/ljspeech/glow-tts --vocoder_name vocoder_models/universal/libri-tts/fullband-melgan
2020-06-08 10:38:55 -04:00
#
# For now, for deployment check the systemd unit in the pull request:
# https://github.com/NixOS/nixpkgs/pull/103851#issue-521121136
#
# Check the latest release for compatible models:
# https://github.com/coqui-ai/TTS/releases/tag/v0.0.11
2020-06-08 10:38:55 -04:00
python3Packages.buildPythonApplication rec {
pname = "tts";
version = "0.0.11";
2020-06-08 10:38:55 -04:00
src = fetchFromGitHub {
owner = "coqui-ai";
2020-06-08 10:38:55 -04:00
repo = "TTS";
rev = "v${version}";
sha256 = "0kk9bgiw2x5ybwz0v3zrfaxajl3lnccc9xmrwc295n2rfkmwxsis";
2020-06-08 10:38:55 -04:00
};
preBuild = ''
# numba jit tries to write to its cache directory
export HOME=$TMPDIR
2021-01-16 08:11:12 -05:00
# we only support pytorch models right now
sed -i -e '/tensorflow/d' requirements.txt
2020-06-08 10:38:55 -04:00
sed -i -e 's!librosa==[^"]*!librosa!' requirements.txt setup.py
sed -i -e 's!unidecode==[^"]*!unidecode!' requirements.txt setup.py
sed -i -e 's!bokeh==[^"]*!bokeh!' requirements.txt setup.py
sed -i -e 's!numba==[^"]*!numba!' requirements.txt setup.py
sed -i -e 's!numpy==[^"]*!numpy!' requirements.txt setup.py
sed -i -e 's!umap-learn==[^"]*!umap-learn!' requirements.txt setup.py
2020-06-08 10:38:55 -04:00
# Not required for building/installation but for their development/ci workflow
2021-01-16 08:11:12 -05:00
sed -i -e '/pylint/d' requirements.txt
2020-06-08 10:38:55 -04:00
sed -i -e '/cardboardlint/d' requirements.txt setup.py
'';
2021-01-16 08:11:12 -05:00
nativeBuildInputs = [ python3Packages.cython ];
2020-06-08 10:38:55 -04:00
propagatedBuildInputs = with python3Packages; [
attrdict
bokeh
flask
2020-06-08 10:38:55 -04:00
fuzzywuzzy
gdown
inflect
jieba
librosa
matplotlib
phonemizer
pypinyin
2020-06-08 10:38:55 -04:00
pysbd
pytorch
scipy
soundfile
tensorboardx
tqdm
umap-learn
unidecode
2020-06-08 10:38:55 -04:00
];
postInstall = ''
cp -r TTS/server/templates/ $out/${python3.sitePackages}/TTS/server
2021-01-16 08:11:12 -05:00
# cython modules are not installed for some reasons
(
cd TTS/tts/layers/glow_tts/monotonic_align
${python3Packages.python.interpreter} setup.py install --prefix=$out
)
2020-06-08 10:38:55 -04:00
'';
checkInputs = with python3Packages; [ pytestCheckHook ];
disabledTests = [
# RuntimeError: fft: ATen not compiled with MKL support
"test_torch_stft"
"test_stft_loss"
"test_multiscale_stft_loss"
# AssertionErrors that I feel incapable of debugging
"test_phoneme_to_sequence"
"test_text2phone"
"test_parametrized_gan_dataset"
];
2021-01-16 08:11:12 -05:00
preCheck = ''
# use the installed TTS in $PYTHONPATH instead of the one from source to also have cython modules.
mv TTS{,.old}
'';
disabledTestPaths = [
2021-01-16 08:11:12 -05:00
# requires tensorflow
"tests/test_tacotron2_tf_model.py"
"tests/test_vocoder_tf_melgan_generator.py"
"tests/test_vocoder_tf_pqmf.py"
2021-01-16 08:11:12 -05:00
];
2020-06-08 10:38:55 -04:00
meta = with lib; {
homepage = "https://github.com/coqui-ai/TTS";
changelog = "https://github.com/coqui-ai/TTS/releases/tag/v${version}";
description = "Deep learning toolkit for Text-to-Speech, battle-tested in research and production";
2020-06-08 10:38:55 -04:00
license = licenses.mpl20;
maintainers = with maintainers; [ hexa mic92 ];
};
}