gador/nixpkgs
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

wyoming-faster-whisper: 2.5.0 -> 3.1.0 (#493887)

Browse Source
This commit is contained in:
Martin Weinelt
2026-02-28 02:28:58 +00:00
committed by GitHub
parent cc7c55e12c 4e807c814a
commit 8ddf32cedd
3 changed files with 292 additions and 234 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nixos/modules/services/home-automation/wyoming/faster-whisper.nix
+262 -222
View File
@@ -18,10 +18,6 @@ let
types
;
inherit (builtins)
toString
;
inherit (utils)
escapeSystemdExecArgs
;
@@ -36,230 +32,261 @@ in
description = ''
Attribute set of wyoming-faster-whisper instances to spawn.
'';
type = attrsOf (submodule {
options = {
enable = mkEnableOption "Wyoming faster-whisper server";
type = attrsOf (
submodule (
{ name, ... }:
{
options = {
enable = mkEnableOption "Wyoming faster-whisper server";
model = mkOption {
type = str;
default = "tiny-int8";
example = "Systran/faster-distil-whisper-small.en";
# https://github.com/home-assistant/addons/blob/master/whisper/DOCS.md#option-model
description = ''
Name of the voice model to use. Can also be a HuggingFace model ID or a path to
a custom model directory.
zeroconf = {
enable = mkEnableOption "zeroconf discovery" // {
default = true;
};
With {option}`useTranformers` enabled, a HuggingFace transformers Whisper model
ID from HuggingFace like `openai/whisper-tiny.en` must be used.
name = mkOption {
type = str;
default = "faster-whisper-${name}";
description = ''
The advertised name for zeroconf discovery.
'';
};
};
Compressed models (`int8`) are slightly less accurate, but smaller and faster.
Distilled models are uncompressed and faster and smaller than non-distilled models.
sttLibrary = mkOption {
type = enum [
"auto"
"faster-whisper"
"onnx-asr"
"sherpa"
"transformers"
];
default = "auto";
example = "sherpa";
description = ''
Library used for speech-to-text process.
Available models:
- `tiny-int8` (compressed)
- `tiny`
- `tiny.en` (English only)
- `base-int8` (compressed)
- `base`
- `base.en` (English only)
- `small-int8` (compressed)
- `distil-small.en` (distilled, English only)
- `small`
- `small.en` (English only)
- `medium-int8` (compressed)
- `distil-medium.en` (distilled, English only)
- `medium`
- `medium.en` (English only)
- `large`
- `large-v1`
- `distil-large-v2` (distilled, English only)
- `large-v2`
- `distil-large-v3` (distilled, English only)
- `large-v3`
- `turbo` (faster than large-v3)
'';
};
When set to `auto` picks a default based on the {option}`language`.
'';
};
useTransformers = mkOption {
type = bool;
default = false;
description = ''
Whether to provide the dependencies to allow using transformer models.
'';
};
model = mkOption {
type = str;
default = "auto";
example = "sherpa-onnx-nemo-parakeet-tdt-0.6b-v3-int8";
# https://github.com/home-assistant/addons/blob/master/whisper/DOCS.md#option-model
description = ''
Name of the voice model to use. Can also be a HuggingFace model ID or a path to
a custom model directory.
uri = mkOption {
type = strMatching "^(tcp|unix)://.*$";
example = "tcp://0.0.0.0:10300";
description = ''
URI to bind the wyoming server to.
'';
};
When set to `auto` picks a default based on the {option}`language`,
device = mkOption {
# https://opennmt.net/CTranslate2/python/ctranslate2.models.Whisper.html#
type = enum [
"cpu"
"cuda"
"auto"
];
default = "cpu";
description = ''
Determines the platform faster-whisper is run on. CPU works everywhere, CUDA requires a compatible NVIDIA GPU.
'';
};
With {option}`sttLibrary` set to `transformers`, a HuggingFace transformers Whisper model
ID from HuggingFace like `openai/whisper-tiny` or `openai/whisper-base` must be used.
language = mkOption {
type = enum [
# https://github.com/home-assistant/addons/blob/master/whisper/config.yaml#L20
"auto"
"af"
"am"
"ar"
"as"
"az"
"ba"
"be"
"bg"
"bn"
"bo"
"br"
"bs"
"ca"
"cs"
"cy"
"da"
"de"
"el"
"en"
"es"
"et"
"eu"
"fa"
"fi"
"fo"
"fr"
"gl"
"gu"
"ha"
"haw"
"he"
"hi"
"hr"
"ht"
"hu"
"hy"
"id"
"is"
"it"
"ja"
"jw"
"ka"
"kk"
"km"
"kn"
"ko"
"la"
"lb"
"ln"
"lo"
"lt"
"lv"
"mg"
"mi"
"mk"
"ml"
"mn"
"mr"
"ms"
"mt"
"my"
"ne"
"nl"
"nn"
"no"
"oc"
"pa"
"pl"
"ps"
"pt"
"ro"
"ru"
"sa"
"sd"
"si"
"sk"
"sl"
"sn"
"so"
"sq"
"sr"
"su"
"sv"
"sw"
"ta"
"te"
"tg"
"th"
"tk"
"tl"
"tr"
"tt"
"uk"
"ur"
"uz"
"vi"
"yi"
"yue"
"yo"
"zh"
];
example = "en";
description = ''
The language used to to parse words and sentences.
'';
};
Compressed models (`int8`) are slightly less accurate, but smaller and faster.
Distilled models are uncompressed and faster and smaller than non-distilled models.
initialPrompt = mkOption {
type = nullOr str;
default = null;
# https://github.com/home-assistant/addons/blob/master/whisper/DOCS.md#option-custom_model_type
example = ''
The following conversation takes place in the universe of
Wizard of Oz. Key terms include 'Yellow Brick Road' (the path
to follow), 'Emerald City' (the ultimate goal), and 'Ruby
Slippers' (the magical tools to succeed). Keep these in mind as
they guide the journey.
'';
description = ''
Optional text to provide as a prompt for the first window. This can be used to provide, or
"prompt-engineer" a context for transcription, e.g. custom vocabularies or proper nouns
to make it more likely to predict those word correctly.
Available models for faster-whisper:
- `tiny-int8` (compressed)
- `tiny`
- `tiny.en` (English only)
- `base-int8` (compressed)
- `base`
- `base.en` (English only)
- `small-int8` (compressed)
- `distil-small.en` (distilled, English only)
- `small`
- `small.en` (English only)
- `medium-int8` (compressed)
- `distil-medium.en` (distilled, English only)
- `medium`
- `medium.en` (English only)
- `large`
- `large-v1`
- `distil-large-v2` (distilled, English only)
- `large-v2`
- `distil-large-v3` (distilled, English only)
- `large-v3`
- `turbo` (faster than large-v3)
'';
};
Not supported when the {option}`customModelType` is `transformers`.
'';
};
uri = mkOption {
type = strMatching "^(tcp|unix)://.*$";
example = "tcp://0.0.0.0:10300";
description = ''
URI to bind the wyoming server to.
'';
};
beamSize = mkOption {
type = ints.unsigned;
default = 0;
example = 5;
description = ''
The number of beams to use in beam search.
Use `0` to automatically select a value based on the CPU.
'';
apply = toString;
};
device = mkOption {
# https://opennmt.net/CTranslate2/python/ctranslate2.models.Whisper.html#
type = enum [
"cpu"
"cuda"
"auto"
];
default = "cpu";
description = ''
Determines the platform faster-whisper is run on. CPU works everywhere, CUDA requires a compatible NVIDIA GPU.
'';
};
extraArgs = mkOption {
type = listOf str;
default = [ ];
description = ''
Extra arguments to pass to the server commandline.
'';
};
};
});
language = mkOption {
type = enum [
# https://github.com/home-assistant/addons/blob/master/whisper/config.yaml#L20
"auto"
"af"
"am"
"ar"
"as"
"az"
"ba"
"be"
"bg"
"bn"
"bo"
"br"
"bs"
"ca"
"cs"
"cy"
"da"
"de"
"el"
"en"
"es"
"et"
"eu"
"fa"
"fi"
"fo"
"fr"
"gl"
"gu"
"ha"
"haw"
"he"
"hi"
"hr"
"ht"
"hu"
"hy"
"id"
"is"
"it"
"ja"
"jw"
"ka"
"kk"
"km"
"kn"
"ko"
"la"
"lb"
"ln"
"lo"
"lt"
"lv"
"mg"
"mi"
"mk"
"ml"
"mn"
"mr"
"ms"
"mt"
"my"
"ne"
"nl"
"nn"
"no"
"oc"
"pa"
"pl"
"ps"
"pt"
"ro"
"ru"
"sa"
"sd"
"si"
"sk"
"sl"
"sn"
"so"
"sq"
"sr"
"su"
"sv"
"sw"
"ta"
"te"
"tg"
"th"
"tk"
"tl"
"tr"
"tt"
"uk"
"ur"
"uz"
"vi"
"yi"
"yue"
"yo"
"zh"
];
example = "en";
description = ''
The language used to to parse words and sentences.
'';
};
initialPrompt = mkOption {
type = nullOr str;
default = null;
# https://github.com/home-assistant/addons/blob/master/whisper/DOCS.md#option-custom_model_type
example = ''
The following conversation takes place in the universe of
Wizard of Oz. Key terms include 'Yellow Brick Road' (the path
to follow), 'Emerald City' (the ultimate goal), and 'Ruby
Slippers' (the magical tools to succeed). Keep these in mind as
they guide the journey.
'';
description = ''
Optional text to provide as a prompt for the first window. This can be used to provide, or
"prompt-engineer" a context for transcription, e.g. custom vocabularies or proper nouns
to make it more likely to predict those word correctly.
Only supported when the {option}`sttLibrary` is `faster-whisper`.
'';
};
beamSize = mkOption {
type = ints.unsigned;
default = 0;
example = 5;
description = ''
The number of beams to use in beam search.
The default (`0`) will use 1 beam on ARM systems (assumes an ARM SBC) and 5 everywhere else.
'';
apply = toString;
};
extraArgs = mkOption {
type = listOf str;
default = [ ];
description = ''
Extra arguments to pass to the server commandline.
'';
};
};
}
)
);
};
};
@@ -273,8 +300,8 @@ in
in
mkIf (cfg.servers != { }) {
assertions = mapAttrsToList (server: options: {
assertion = options.useTransformers -> options.initialPrompt == null;
message = "wyoming-faster-whisper/${server}: Transformer models (`useTransformers`) do not currently support an `initialPrompt`.";
assertion = options.sttLibrary != "faster-whisper" -> options.initialPrompt == null;
message = "wyoming-faster-whisper/${server}: Initial prompt is only supported when using `faster-whisper` as `sttLibrary`.";
}) cfg.servers;
systemd.services = mapAttrs' (
@@ -283,8 +310,14 @@ in
finalPackage = cfg.package.overridePythonAttrs (oldAttrs: {
dependencies =
oldAttrs.dependencies
# for transformer model support
++ optionals options.useTransformers oldAttrs.optional-dependencies.transformers;
++ optionals options.zeroconf.enable oldAttrs.optional-dependencies.zeroconf
++ optionals (
options.sttLibrary == "onnx-asr" || options.sttLibrary == "auto" && options.language == "ru"
) oldAttrs.optional-dependencies.onnx_asr
++ optionals (
options.sttLibrary == "sherpa-onnx" || options.sttLibrary == "auto" && options.language == "en"
) oldAttrs.optional-dependencies.sherpa
++ optionals (options.sttLibrary == "transformers") oldAttrs.optional-dependencies.transformers;
});
in
nameValuePair "wyoming-faster-whisper-${server}" {
@@ -316,6 +349,8 @@ in
options.uri
"--device"
options.device
"--stt-library"
options.sttLibrary
"--model"
options.model
"--language"
@@ -323,8 +358,9 @@ in
"--beam-size"
options.beamSize
]
++ lib.optionals options.useTransformers [
"--use-transformers"
++ lib.optionals options.zeroconf.enable [
"--zeroconf"
options.zeroconf.name
]
++ lib.optionals (options.initialPrompt != null) [
"--initial-prompt"
@@ -366,6 +402,10 @@ in
"AF_INET"
"AF_INET6"
"AF_UNIX"
]
++ lib.optionals options.zeroconf.enable [
# Zeroconf support require network interface enumeration
"AF_NETLINK"
];
RestrictNamespaces = true;
RestrictRealtime = true;
nixos/modules/services/home-automation/wyoming/piper.nix
+6 -3
View File
@@ -37,7 +37,7 @@ in
'';
type = types.attrsOf (
types.submodule (
{ ... }:
{ name, ... }:
{
options = {
enable = mkEnableOption "Wyoming Piper server";
@@ -49,7 +49,7 @@ in
name = mkOption {
type = str;
default = "piper";
default = "piper-${name}";
description = ''
The advertised name for zeroconf discovery.
'';
@@ -205,8 +205,11 @@ in
RestrictAddressFamilies = [
"AF_INET"
"AF_INET6"
"AF_NETLINK"
"AF_UNIX"
]
++ lib.optionals options.zeroconf.enable [
# Zeroconf support require network interface enumeration
"AF_NETLINK"
];
RestrictNamespaces = true;
RestrictRealtime = true;
pkgs/by-name/wy/wyoming-faster-whisper/package.nix
+24 -9
View File
@@ -6,14 +6,14 @@
python3Packages.buildPythonApplication (finalAttrs: {
pname = "wyoming-faster-whisper";
version = "2.5.0";
version = "3.1.0";
pyproject = true;
src = fetchFromGitHub {
owner = "rhasspy";
repo = "wyoming-faster-whisper";
tag = "v${finalAttrs.version}";
hash = "sha256-MKB6gZdGdAYoNK8SRiDHG8xtMZ5mXdaSn+bH4T6o/K4=";
hash = "sha256-p1FCyj/D7ndKJD1/V5YzhT0xlkg61DSx2m3DCELmPO8=";
};
build-system = with python3Packages; [
@@ -25,18 +25,33 @@ python3Packages.buildPythonApplication (finalAttrs: {
"wyoming"
];
pythonRemoveDeps = [
# https://github.com/rhasspy/wyoming-faster-whisper/pull/81
"requests"
];
dependencies = with python3Packages; [
faster-whisper
wyoming
];
optional-dependencies = {
transformers =
with python3Packages;
[
transformers
]
++ transformers.optional-dependencies.torch;
optional-dependencies = with python3Packages; {
transformers = [
transformers
]
++ transformers.optional-dependencies.torch;
sherpa = [
sherpa-onnx
];
onnx_asr = [
onnx-asr
]
++ onnx-asr.optional-dependencies.cpu
++ onnx-asr.optional-dependencies.hub;
zeroconf = [
wyoming
]
++ wyoming.optional-dependencies.zeroconf;
};
pythonImportsCheck = [