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

126 lines
3.3 KiB
Nix
Raw Normal View History

2020-06-08 15:38:55 +01:00
{ lib
2021-07-26 19:41:47 +01:00
, python3
2020-06-08 15:38:55 +01:00
, fetchFromGitHub
2021-04-16 08:33:04 +01:00
, fetchpatch
2020-06-08 15:38:55 +01:00
}:
2021-01-16 13:11:12 +00:00
# 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 15:38:55 +01:00
#
2021-04-16 08:33:04 +01:00
# If you upgrade from an old version you may have to delete old models from ~/.local/share/tts
#
2020-06-08 15:38:55 +01:00
# For now, for deployment check the systemd unit in the pull request:
# https://github.com/NixOS/nixpkgs/pull/103851#issue-521121136
2021-07-26 19:41:47 +01:00
python3.pkgs.buildPythonApplication rec {
2020-06-08 15:38:55 +01:00
pname = "tts";
2021-08-31 13:26:32 +01:00
version = "0.2.1";
2021-07-02 13:00:46 +01:00
2020-06-08 15:38:55 +01:00
src = fetchFromGitHub {
owner = "coqui-ai";
2020-06-08 15:38:55 +01:00
repo = "TTS";
rev = "v${version}";
2021-08-31 13:26:32 +01:00
sha256 = "sha256-7YMNxZ15qQowEE0tE6x/LbtirNGp7h9OLyS1JSl9x2A=";
2020-06-08 15:38:55 +01:00
};
postPatch = ''
2021-08-28 02:16:17 +01:00
sed -i requirements.txt \
-e 's!librosa==[^"]*!librosa!' \
-e 's!mecab-python3==[^"]*!mecab-python3!' \
-e 's!numba==[^"]*!numba!' \
-e 's!numpy==[^"]*!numpy!' \
-e 's!umap-learn==[^"]*!umap-learn!'
2020-06-08 15:38:55 +01:00
'';
2021-07-02 13:00:46 +01:00
nativeBuildInputs = with python3.pkgs; [
2021-05-14 19:26:52 +01:00
cython
];
2020-06-08 15:38:55 +01:00
2021-07-02 13:00:46 +01:00
propagatedBuildInputs = with python3.pkgs; [
2021-06-04 13:46:55 +01:00
anyascii
coqpit
flask
2021-08-11 13:53:10 +01:00
fsspec
2021-07-03 20:31:31 +01:00
gruut
2020-06-08 15:38:55 +01:00
gdown
inflect
jieba
librosa
matplotlib
2021-06-04 13:46:55 +01:00
mecab-python3
numba
2021-05-14 19:26:52 +01:00
pandas
pypinyin
2020-06-08 15:38:55 +01:00
pysbd
pytorch
scipy
soundfile
tensorboardx
2021-06-04 13:46:55 +01:00
tensorflow
tqdm
umap-learn
2021-06-04 13:46:55 +01:00
unidic-lite
2020-06-08 15:38:55 +01:00
];
postInstall = ''
cp -r TTS/server/templates/ $out/${python3.sitePackages}/TTS/server
2021-01-16 13:11:12 +00:00
# cython modules are not installed for some reasons
(
cd TTS/tts/layers/glow_tts/monotonic_align
2021-07-02 13:00:46 +01:00
${python3.interpreter} setup.py install --prefix=$out
2021-01-16 13:11:12 +00:00
)
2020-06-08 15:38:55 +01:00
'';
2021-07-02 13:00:46 +01:00
checkInputs = with python3.pkgs; [
pytest-sugar
2021-05-14 19:26:52 +01:00
pytestCheckHook
];
2020-06-08 15:38:55 +01:00
disabledTests = [
# RuntimeError: fft: ATen not compiled with MKL support
"test_torch_stft"
"test_stft_loss"
"test_multiscale_stft_loss"
2021-05-14 19:26:52 +01:00
# Requires network acccess to download models
"test_synthesize"
2020-06-08 15:38:55 +01:00
];
2021-01-16 13:11:12 +00:00
preCheck = ''
# use the installed TTS in $PYTHONPATH instead of the one from source to also have cython modules.
mv TTS{,.old}
2021-05-14 19:26:52 +01:00
export PATH=$out/bin:$PATH
# numba tries to write to HOME directory
export HOME=$TMPDIR
for file in $(grep -rl 'python TTS/bin' tests); do
substituteInPlace "$file" \
--replace "python TTS/bin" "${python3.interpreter} $out/lib/${python3.libPrefix}/site-packages/TTS/bin"
done
2021-01-16 13:11:12 +00:00
'';
disabledTestPaths = [
2021-01-16 13:11:12 +00:00
# requires tensorflow
"tests/vocoder_tests/test_vocoder_tf_pqmf.py"
"tests/vocoder_tests/test_vocoder_tf_melgan_generator.py"
2021-06-04 13:46:55 +01:00
"tests/tts_tests/test_tacotron2_tf_model.py"
# RuntimeError: fft: ATen not compiled with MKL support
2021-08-11 13:53:10 +01:00
"tests/tts_tests/test_vits_train.py"
"tests/vocoder_tests/test_fullband_melgan_train.py"
"tests/vocoder_tests/test_hifigan_train.py"
"tests/vocoder_tests/test_melgan_train.py"
"tests/vocoder_tests/test_multiband_melgan_train.py"
"tests/vocoder_tests/test_parallel_wavegan_train.py"
2021-01-16 13:11:12 +00:00
];
2020-06-08 15:38:55 +01:00
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 15:38:55 +01:00
license = licenses.mpl20;
maintainers = with maintainers; [ hexa mic92 ];
};
}