From 438f14fbc9e63e95a6430213946481828049da93 Mon Sep 17 00:00:00 2001
From: Ryo Yamashita <qryxip@gmail.com>
Date: Sat, 7 Dec 2024 15:31:40 +0900
Subject: [PATCH 1/2] =?UTF-8?q?feat!:=20`RunAsync`=E3=82=92=E4=BD=BF?=
=?UTF-8?q?=E3=81=86?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
Cargo.lock | 4 +-
Cargo.toml | 2 +-
crates/voicevox_core/src/infer.rs | 58 +++++--
.../src/infer/runtimes/onnxruntime.rs | 86 +++++-----
crates/voicevox_core/src/infer/session_set.rs | 19 ++-
crates/voicevox_core/src/status.rs | 6 +-
crates/voicevox_core/src/synthesizer.rs | 153 +++++++++---------
.../src/inference_domain.rs | 16 +-
.../test/test_asyncio_user_dict_load.py | 9 +-
example/python/run-asyncio.py | 8 +-
example/python/run.py | 8 +-
11 files changed, 225 insertions(+), 144 deletions(-)
diff --git a/Cargo.lock b/Cargo.lock
index 2005f6ffa..c187c0cf1 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -4251,7 +4251,7 @@ checksum = "49874b5167b65d7193b8aba1567f5c7d93d001cafc34600cee003eda787e483f"
[[package]]
name = "voicevox-ort"
version = "2.0.0-rc.4"
-source = "git+https://github.com/VOICEVOX/ort.git?rev=17f741301db0bb08da0eafe8a338e5efd8a4b5df#17f741301db0bb08da0eafe8a338e5efd8a4b5df"
+source = "git+https://github.com/VOICEVOX/ort.git?rev=09a9fe1619c1561efafc02f68f0bda4aad879771#09a9fe1619c1561efafc02f68f0bda4aad879771"
dependencies = [
"anyhow",
"half",
@@ -4268,7 +4268,7 @@ dependencies = [
[[package]]
name = "voicevox-ort-sys"
version = "2.0.0-rc.4"
-source = "git+https://github.com/VOICEVOX/ort.git?rev=17f741301db0bb08da0eafe8a338e5efd8a4b5df#17f741301db0bb08da0eafe8a338e5efd8a4b5df"
+source = "git+https://github.com/VOICEVOX/ort.git?rev=09a9fe1619c1561efafc02f68f0bda4aad879771#09a9fe1619c1561efafc02f68f0bda4aad879771"
dependencies = [
"flate2",
"pkg-config",
diff --git a/Cargo.toml b/Cargo.toml
index a3a606317..3c3c184c5 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -99,7 +99,7 @@ zip = "0.6.3"
[workspace.dependencies.voicevox-ort]
git = "https://github.com/VOICEVOX/ort.git"
-rev = "17f741301db0bb08da0eafe8a338e5efd8a4b5df"
+rev = "09a9fe1619c1561efafc02f68f0bda4aad879771"
[workspace.dependencies.open_jtalk]
git = "https://github.com/VOICEVOX/open_jtalk-rs.git"
diff --git a/crates/voicevox_core/src/infer.rs b/crates/voicevox_core/src/infer.rs
index e827ddd7d..16b246e36 100644
--- a/crates/voicevox_core/src/infer.rs
+++ b/crates/voicevox_core/src/infer.rs
@@ -3,7 +3,7 @@ mod model_file;
pub(crate) mod runtimes;
pub(crate) mod session_set;
-use std::{borrow::Cow, collections::BTreeSet, fmt::Debug, ops::Index, sync::Arc};
+use std::{borrow::Cow, collections::BTreeSet, fmt::Debug, future::Future, ops::Index, sync::Arc};
use derive_new::new;
use duplicate::duplicate_item;
@@ -12,14 +12,39 @@ use ndarray::{Array, ArrayD, Dimension, ShapeError};
use thiserror::Error;
use crate::{
+ asyncs::{Async, BlockingThreadPool, SingleTasked},
devices::{DeviceSpec, GpuSpec},
StyleType, SupportedDevices,
};
+pub(crate) trait AsyncExt: Async {
+ async fn run_session<R: InferenceRuntime>(
+ ctx: R::RunContext,
+ ) -> anyhow::Result<Vec<OutputTensor>>;
+}
+
+impl AsyncExt for SingleTasked {
+ async fn run_session<R: InferenceRuntime>(
+ ctx: R::RunContext,
+ ) -> anyhow::Result<Vec<OutputTensor>> {
+ R::run(ctx)
+ }
+}
+
+impl AsyncExt for BlockingThreadPool {
+ async fn run_session<R: InferenceRuntime>(
+ ctx: R::RunContext,
+ ) -> anyhow::Result<Vec<OutputTensor>> {
+ R::run_async(ctx).await
+ }
+}
+
pub(crate) trait InferenceRuntime: 'static {
// TODO: "session"とは何なのかを定め、ドキュメントを書く。`InferenceSessionSet`も同様。
- type Session: Sized + Send + 'static;
- type RunContext<'a>: From<&'a mut Self::Session> + PushInputTensor;
+ type Session;
+
+ // 本当は`From<'_ Self::Session>`としたいが、 rust-lang/rust#100013 がある
+ type RunContext: From<Arc<Self::Session>> + PushInputTensor;
/// 名前。
const DISPLAY_NAME: &'static str;
@@ -45,7 +70,11 @@ pub(crate) trait InferenceRuntime: 'static {
Vec<ParamInfo<OutputScalarKind>>,
)>;
- fn run(ctx: Self::RunContext<'_>) -> anyhow::Result<Vec<OutputTensor>>;
+ fn run(ctx: Self::RunContext) -> anyhow::Result<Vec<OutputTensor>>;
+
+ fn run_async(
+ ctx: Self::RunContext,
+ ) -> impl Future<Output = anyhow::Result<Vec<OutputTensor>>> + Send;
}
/// 共に扱われるべき推論操作の集合を示す。
@@ -101,8 +130,8 @@ pub(crate) trait InferenceInputSignature: Send + 'static {
const PARAM_INFOS: &'static [ParamInfo<InputScalarKind>];
fn make_run_context<R: InferenceRuntime>(
self,
- sess: &mut R::Session,
- ) -> anyhow::Result<R::RunContext<'_>>;
+ sess: Arc<R::Session>,
+ ) -> anyhow::Result<R::RunContext>;
}
pub(crate) trait InputScalar: Sized {
@@ -110,6 +139,7 @@ pub(crate) trait InputScalar: Sized {
// TODO: `Array`ではなく`ArrayView`を取ることができるかもしれない
fn push_tensor_to_ctx(
+ name: &'static str,
tensor: Array<Self, impl Dimension + 'static>,
visitor: &mut impl PushInputTensor,
) -> anyhow::Result<()>;
@@ -124,10 +154,11 @@ impl InputScalar for T {
const KIND: InputScalarKind = KIND_VAL;
fn push_tensor_to_ctx(
+ name: &'static str,
tensor: Array<Self, impl Dimension + 'static>,
ctx: &mut impl PushInputTensor,
) -> anyhow::Result<()> {
- ctx.push(tensor)
+ ctx.push(name, tensor)
}
}
@@ -141,8 +172,17 @@ pub(crate) enum InputScalarKind {
}
pub(crate) trait PushInputTensor {
- fn push_int64(&mut self, tensor: Array<i64, impl Dimension + 'static>) -> anyhow::Result<()>;
- fn push_float32(&mut self, tensor: Array<f32, impl Dimension + 'static>) -> anyhow::Result<()>;
+ fn push_int64(
+ &mut self,
+ name: &'static str,
+ tensor: Array<i64, impl Dimension + 'static>,
+ ) -> anyhow::Result<()>;
+
+ fn push_float32(
+ &mut self,
+ name: &'static str,
+ tensor: Array<f32, impl Dimension + 'static>,
+ ) -> anyhow::Result<()>;
}
/// 推論操作の出力シグネチャ。
diff --git a/crates/voicevox_core/src/infer/runtimes/onnxruntime.rs b/crates/voicevox_core/src/infer/runtimes/onnxruntime.rs
index ffeb55e5f..8e457bbf0 100644
--- a/crates/voicevox_core/src/infer/runtimes/onnxruntime.rs
+++ b/crates/voicevox_core/src/infer/runtimes/onnxruntime.rs
@@ -11,7 +11,7 @@
// }
// ```
-use std::{fmt::Debug, vec};
+use std::{fmt::Debug, sync::Arc, vec};
use anyhow::{anyhow, bail, ensure};
use duplicate::duplicate_item;
@@ -32,8 +32,8 @@ use super::super::{
};
impl InferenceRuntime for self::blocking::Onnxruntime {
- type Session = ort::Session;
- type RunContext<'a> = OnnxruntimeRunContext<'a>;
+ type Session = async_lock::Mutex<ort::Session>; // WASMでは`ort`を利用しないので、ここはasync-lockを用いてよいはず
+ type RunContext = OnnxruntimeRunContext;
const DISPLAY_NAME: &'static str = if cfg!(feature = "load-onnxruntime") {
"現在ロードされているONNX Runtime"
@@ -179,58 +179,44 @@ impl InferenceRuntime for self::blocking::Onnxruntime {
})
.collect::<anyhow::Result<_>>()?;
- Ok((sess, input_param_infos, output_param_infos))
+ Ok((sess.into(), input_param_infos, output_param_infos))
}
fn run(
- OnnxruntimeRunContext { sess, inputs }: OnnxruntimeRunContext<'_>,
+ OnnxruntimeRunContext { sess, inputs }: Self::RunContext,
) -> anyhow::Result<Vec<OutputTensor>> {
- let outputs = sess.run(&*inputs)?;
-
- (0..outputs.len())
- .map(|i| {
- let output = &outputs[i];
-
- let ValueType::Tensor { ty, .. } = output.dtype()? else {
- bail!(
- "unexpected output. currently `ONNX_TYPE_TENSOR` and \
- `ONNX_TYPE_SPARSETENSOR` is supported",
- );
- };
+ extract_outputs(&sess.lock_blocking().run(inputs)?)
+ }
- match ty {
- TensorElementType::Float32 => {
- let output = output.try_extract_tensor::<f32>()?;
- Ok(OutputTensor::Float32(output.into_owned()))
- }
- _ => bail!("unexpected output tensor element data type"),
- }
- })
- .collect()
+ async fn run_async(
+ OnnxruntimeRunContext { sess, inputs }: Self::RunContext,
+ ) -> anyhow::Result<Vec<OutputTensor>> {
+ extract_outputs(&sess.lock().await.run_async(inputs)?.await?)
}
}
-pub(crate) struct OnnxruntimeRunContext<'sess> {
- sess: &'sess ort::Session,
- inputs: Vec<ort::SessionInputValue<'static>>,
+pub(crate) struct OnnxruntimeRunContext {
+ sess: Arc<async_lock::Mutex<ort::Session>>,
+ inputs: Vec<(&'static str, ort::SessionInputValue<'static>)>,
}
-impl OnnxruntimeRunContext<'_> {
+impl OnnxruntimeRunContext {
fn push_input(
&mut self,
+ name: &'static str,
input: Array<
impl PrimitiveTensorElementType + Debug + Clone + 'static,
impl Dimension + 'static,
>,
) -> anyhow::Result<()> {
let input = ort::Value::from_array(input)?.into();
- self.inputs.push(input);
+ self.inputs.push((name, input));
Ok(())
}
}
-impl<'sess> From<&'sess mut ort::Session> for OnnxruntimeRunContext<'sess> {
- fn from(sess: &'sess mut ort::Session) -> Self {
+impl From<Arc<async_lock::Mutex<ort::Session>>> for OnnxruntimeRunContext {
+ fn from(sess: Arc<async_lock::Mutex<ort::Session>>) -> Self {
Self {
sess,
inputs: vec![],
@@ -238,17 +224,45 @@ impl<'sess> From<&'sess mut ort::Session> for OnnxruntimeRunContext<'sess> {
}
}
-impl PushInputTensor for OnnxruntimeRunContext<'_> {
+impl PushInputTensor for OnnxruntimeRunContext {
#[duplicate_item(
method T;
[ push_int64 ] [ i64 ];
[ push_float32 ] [ f32 ];
)]
- fn method(&mut self, tensor: Array<T, impl Dimension + 'static>) -> anyhow::Result<()> {
- self.push_input(tensor)
+ fn method(
+ &mut self,
+ name: &'static str,
+ tensor: Array<T, impl Dimension + 'static>,
+ ) -> anyhow::Result<()> {
+ self.push_input(name, tensor)
}
}
+// FIXME: use ouroboros to reduce copies
+fn extract_outputs(outputs: &ort::SessionOutputs<'_, '_>) -> anyhow::Result<Vec<OutputTensor>> {
+ (0..outputs.len())
+ .map(|i| {
+ let output = &outputs[i];
+
+ let ValueType::Tensor { ty, .. } = output.dtype()? else {
+ bail!(
+ "unexpected output. currently `ONNX_TYPE_TENSOR` and `ONNX_TYPE_SPARSETENSOR`
+ is supported",
+ );
+ };
+
+ match ty {
+ TensorElementType::Float32 => {
+ let output = output.try_extract_tensor::<f32>()?;
+ Ok(OutputTensor::Float32(output.into_owned()))
+ }
+ _ => bail!("unexpected output tensor element data type"),
+ }
+ })
+ .collect()
+}
+
pub(crate) mod blocking {
use ort::EnvHandle;
use ref_cast::{ref_cast_custom, RefCastCustom};
diff --git a/crates/voicevox_core/src/infer/session_set.rs b/crates/voicevox_core/src/infer/session_set.rs
index e94fff962..1bac459b7 100644
--- a/crates/voicevox_core/src/infer/session_set.rs
+++ b/crates/voicevox_core/src/infer/session_set.rs
@@ -12,7 +12,7 @@ use super::{
};
pub(crate) struct InferenceSessionSet<R: InferenceRuntime, D: InferenceDomain>(
- EnumMap<D::Operation, Arc<std::sync::Mutex<R::Session>>>,
+ EnumMap<D::Operation, Arc<R::Session>>,
);
impl<R: InferenceRuntime, D: InferenceDomain> InferenceSessionSet<R, D> {
@@ -33,7 +33,7 @@ impl<R: InferenceRuntime, D: InferenceDomain> InferenceSessionSet<R, D> {
check_param_infos(expected_input_param_infos, &actual_input_param_infos)?;
check_param_infos(expected_output_param_infos, &actual_output_param_infos)?;
- Ok((op.into_usize(), std::sync::Mutex::new(sess).into()))
+ Ok((op.into_usize(), sess.into()))
})
.collect::<anyhow::Result<HashMap<_, _>>>()?;
@@ -84,18 +84,21 @@ impl<R: InferenceRuntime, D: InferenceDomain> InferenceSessionSet<R, D> {
}
pub(crate) struct InferenceSessionCell<R: InferenceRuntime, I> {
- inner: Arc<std::sync::Mutex<R::Session>>,
+ inner: Arc<R::Session>,
marker: PhantomData<fn(I)>,
}
impl<R: InferenceRuntime, I: InferenceInputSignature> InferenceSessionCell<R, I> {
- pub(crate) fn run(
+ pub(crate) async fn run<A: super::AsyncExt>(
self,
input: I,
) -> crate::Result<<I::Signature as InferenceSignature>::Output> {
- let inner = &mut self.inner.lock().unwrap();
- (|| R::run(input.make_run_context::<R>(inner)?)?.try_into())()
- .map_err(ErrorRepr::RunModel)
- .map_err(Into::into)
+ async {
+ let ctx = input.make_run_context::<R>(self.inner.clone())?;
+ A::run_session::<R>(ctx).await?.try_into()
+ }
+ .await
+ .map_err(ErrorRepr::RunModel)
+ .map_err(Into::into)
}
}
diff --git a/crates/voicevox_core/src/status.rs b/crates/voicevox_core/src/status.rs
index 64986f627..c59573412 100644
--- a/crates/voicevox_core/src/status.rs
+++ b/crates/voicevox_core/src/status.rs
@@ -9,6 +9,7 @@ use itertools::iproduct;
use crate::{
error::{ErrorRepr, LoadModelError, LoadModelErrorKind, LoadModelResult},
infer::{
+ self,
domains::{inference_domain_map_values, InferenceDomainMap, TalkDomain},
session_set::{InferenceSessionCell, InferenceSessionSet},
InferenceDomain, InferenceInputSignature, InferenceRuntime, InferenceSessionOptions,
@@ -104,17 +105,18 @@ impl<R: InferenceRuntime> Status<R> {
/// # Panics
///
/// `self`が`model_id`を含んでいないとき、パニックする。
- pub(crate) fn run_session<I>(
+ pub(crate) async fn run_session<A, I>(
&self,
model_id: VoiceModelId,
input: I,
) -> Result<<I::Signature as InferenceSignature>::Output>
where
+ A: infer::AsyncExt,
I: InferenceInputSignature,
<I::Signature as InferenceSignature>::Domain: InferenceDomainExt,
{
let sess = self.loaded_models.lock().unwrap().get(model_id);
- sess.run(input)
+ sess.run::<A>(input).await
}
}
diff --git a/crates/voicevox_core/src/synthesizer.rs b/crates/voicevox_core/src/synthesizer.rs
index 31aec6b72..230a26294 100644
--- a/crates/voicevox_core/src/synthesizer.rs
+++ b/crates/voicevox_core/src/synthesizer.rs
@@ -1,4 +1,7 @@
-use crate::asyncs::{Async, BlockingThreadPool, SingleTasked};
+use crate::{
+ asyncs::{BlockingThreadPool, SingleTasked},
+ infer,
+};
pub use self::inner::MARGIN;
@@ -70,14 +73,14 @@ pub struct InitializeOptions {
pub cpu_num_threads: u16,
}
-trait AsyncForOnnxruntime: Async {
+trait AsyncExt: infer::AsyncExt {
async fn unblock<T, F>(f: F) -> T
where
F: FnOnce() -> T + Send + 'static,
T: Send + 'static;
}
-impl AsyncForOnnxruntime for SingleTasked {
+impl AsyncExt for SingleTasked {
async fn unblock<T, F>(f: F) -> T
where
F: FnOnce() -> T + Send + 'static,
@@ -87,7 +90,7 @@ impl AsyncForOnnxruntime for SingleTasked {
}
}
-impl AsyncForOnnxruntime for BlockingThreadPool {
+impl AsyncExt for BlockingThreadPool {
async fn unblock<T, F>(f: F) -> T
where
F: FnOnce() -> T + Send + 'static,
@@ -108,11 +111,12 @@ mod inner {
use tracing::info;
use crate::{
- asyncs::{BlockingThreadPool, SingleTasked},
+ asyncs::{Async, BlockingThreadPool, SingleTasked},
devices::{DeviceSpec, GpuSpec},
engine::{create_kana, mora_to_text, wav_from_s16le, Mora, OjtPhoneme},
error::ErrorRepr,
infer::{
+ self,
domains::{
GenerateFullIntermediateInput, GenerateFullIntermediateOutput, InferenceDomainMap,
PredictDurationInput, PredictDurationOutput, PredictIntonationInput,
@@ -127,7 +131,7 @@ mod inner {
SynthesisOptions, VoiceModelId, VoiceModelMeta,
};
- use super::{AccelerationMode, AsyncForOnnxruntime, InitializeOptions, TtsOptions};
+ use super::{AccelerationMode, AsyncExt, InitializeOptions, TtsOptions};
const DEFAULT_SAMPLING_RATE: u32 = 24000;
/// 音が途切れてしまうのを避けるworkaround処理のためのパディング幅（フレーム数）
@@ -169,7 +173,7 @@ mod inner {
audio_query: AudioQuery,
}
- pub struct Inner<O, A> {
+ pub struct Inner<O, A: Async> {
status: Arc<Status<crate::blocking::Onnxruntime>>,
open_jtalk_analyzer: OpenJTalkAnalyzer<O>,
kana_analyzer: KanaAnalyzer,
@@ -189,7 +193,7 @@ mod inner {
}
}
- impl<O, A: AsyncForOnnxruntime> Inner<O, A> {
+ impl<O, A: AsyncExt> Inner<O, A> {
pub(super) fn new(
onnxruntime: &'static crate::blocking::Onnxruntime,
open_jtalk: O,
@@ -751,7 +755,7 @@ mod inner {
}
}
- impl<O: FullcontextExtractor, A: AsyncForOnnxruntime> Inner<O, A> {
+ impl<O: FullcontextExtractor, A: AsyncExt> Inner<O, A> {
pub(super) async fn create_accent_phrases(
&self,
text: &str,
@@ -782,7 +786,8 @@ mod inner {
}
}
- impl<O, A: AsyncForOnnxruntime> Inner<O, A> {
+ // TODO: この層を破壊する
+ impl<O, A: infer::AsyncExt> Inner<O, A> {
pub(super) async fn predict_duration(
&self,
phoneme_vector: &[i64],
@@ -790,7 +795,7 @@ mod inner {
) -> Result<Vec<f32>> {
let status = self.status.clone();
let phoneme_vector = ndarray::arr1(phoneme_vector);
- A::unblock(move || status.predict_duration(phoneme_vector, style_id)).await
+ status.predict_duration::<A>(phoneme_vector, style_id).await
}
#[expect(
@@ -816,8 +821,8 @@ mod inner {
let end_accent_vector = ndarray::arr1(end_accent_vector);
let start_accent_phrase_vector = ndarray::arr1(start_accent_phrase_vector);
let end_accent_phrase_vector = ndarray::arr1(end_accent_phrase_vector);
- A::unblock(move || {
- status.predict_intonation(
+ status
+ .predict_intonation::<A>(
length,
vowel_phoneme_vector,
consonant_phoneme_vector,
@@ -827,8 +832,7 @@ mod inner {
end_accent_phrase_vector,
style_id,
)
- })
- .await
+ .await
}
pub(super) async fn generate_full_intermediate(
@@ -842,16 +846,9 @@ mod inner {
let status = self.status.clone();
let f0 = ndarray::arr1(f0);
let phoneme_vector = ndarray::arr1(phoneme_vector);
- A::unblock(move || {
- status.generate_full_intermediate(
- length,
- phoneme_size,
- f0,
- phoneme_vector,
- style_id,
- )
- })
- .await
+ status
+ .generate_full_intermediate::<A>(length, phoneme_size, f0, phoneme_vector, style_id)
+ .await
}
pub(super) async fn render_audio_segment(
@@ -860,7 +857,7 @@ mod inner {
style_id: StyleId,
) -> Result<ndarray::Array1<f32>> {
let status = self.status.clone();
- A::unblock(move || status.render_audio_segment(spec, style_id)).await
+ status.render_audio_segment::<A>(spec, style_id).await
}
pub(super) async fn decode(
@@ -874,14 +871,14 @@ mod inner {
let status = self.status.clone();
let f0 = ndarray::arr1(f0);
let phoneme_vector = ndarray::arr1(phoneme_vector);
- A::unblock(move || status.decode(length, phoneme_size, f0, phoneme_vector, style_id))
+ status
+ .decode::<A>(length, phoneme_size, f0, phoneme_vector, style_id)
.await
}
}
impl<R: InferenceRuntime> Status<R> {
- /// CPU-boundな操作なので、非同期ランタイム上では直接実行されるべきではない。
- fn predict_duration(
+ pub(super) async fn predict_duration<A: infer::AsyncExt>(
&self,
phoneme_vector: ndarray::Array1<i64>,
style_id: StyleId,
@@ -890,13 +887,15 @@ mod inner {
let PredictDurationOutput {
phoneme_length: output,
- } = self.run_session(
- model_id,
- PredictDurationInput {
- phoneme_list: phoneme_vector,
- speaker_id: ndarray::arr1(&[inner_voice_id.raw_id().into()]),
- },
- )?;
+ } = self
+ .run_session::<A, _>(
+ model_id,
+ PredictDurationInput {
+ phoneme_list: phoneme_vector,
+ speaker_id: ndarray::arr1(&[inner_voice_id.raw_id().into()]),
+ },
+ )
+ .await?;
let mut output = output.into_raw_vec();
for output_item in output.iter_mut() {
@@ -910,13 +909,12 @@ mod inner {
const PHONEME_LENGTH_MINIMAL: f32 = 0.01;
}
- /// CPU-boundな操作なので、非同期ランタイム上では直接実行されるべきではない。
#[expect(
clippy::too_many_arguments,
reason = "compatible_engineでの`predict_intonation`の形を考えると、ここの引数を構造体に\
まとめたりしても可読性に寄与しない"
)]
- fn predict_intonation(
+ pub(super) async fn predict_intonation<A: infer::AsyncExt>(
&self,
length: usize,
vowel_phoneme_vector: ndarray::Array1<i64>,
@@ -929,19 +927,21 @@ mod inner {
) -> Result<Vec<f32>> {
let (model_id, inner_voice_id) = self.ids_for::<TalkDomain>(style_id)?;
- let PredictIntonationOutput { f0_list: output } = self.run_session(
- model_id,
- PredictIntonationInput {
- length: ndarray::arr0(length as i64),
- vowel_phoneme_list: vowel_phoneme_vector,
- consonant_phoneme_list: consonant_phoneme_vector,
- start_accent_list: start_accent_vector,
- end_accent_list: end_accent_vector,
- start_accent_phrase_list: start_accent_phrase_vector,
- end_accent_phrase_list: end_accent_phrase_vector,
- speaker_id: ndarray::arr1(&[inner_voice_id.raw_id().into()]),
- },
- )?;
+ let PredictIntonationOutput { f0_list: output } = self
+ .run_session::<A, _>(
+ model_id,
+ PredictIntonationInput {
+ length: ndarray::arr0(length as i64),
+ vowel_phoneme_list: vowel_phoneme_vector,
+ consonant_phoneme_list: consonant_phoneme_vector,
+ start_accent_list: start_accent_vector,
+ end_accent_list: end_accent_vector,
+ start_accent_phrase_list: start_accent_phrase_vector,
+ end_accent_phrase_list: end_accent_phrase_vector,
+ speaker_id: ndarray::arr1(&[inner_voice_id.raw_id().into()]),
+ },
+ )
+ .await?;
Ok(output.into_raw_vec())
}
@@ -949,9 +949,7 @@ mod inner {
/// モデル`generate_full_intermediate`の実行と、その前後の処理を行う。
///
/// 無音パディングを付加して音声特徴量を計算し、マージン込みの音声特徴量を返す。
- ///
- /// CPU-boundな操作なので、非同期ランタイム上では直接実行されるべきではない。
- fn generate_full_intermediate(
+ pub(super) async fn generate_full_intermediate<A: infer::AsyncExt>(
&self,
length: usize,
phoneme_size: usize,
@@ -973,16 +971,18 @@ mod inner {
let GenerateFullIntermediateOutput {
spec: spec_with_padding,
- } = self.run_session(
- model_id,
- GenerateFullIntermediateInput {
- f0: f0_with_padding
- .into_shape([length_with_padding, 1])
- .unwrap(),
- phoneme: phoneme_with_padding,
- speaker_id: ndarray::arr1(&[inner_voice_id.raw_id().into()]),
- },
- )?;
+ } = self
+ .run_session::<A, _>(
+ model_id,
+ GenerateFullIntermediateInput {
+ f0: f0_with_padding
+ .into_shape([length_with_padding, 1])
+ .unwrap(),
+ phoneme: phoneme_with_padding,
+ speaker_id: ndarray::arr1(&[inner_voice_id.raw_id().into()]),
+ },
+ )
+ .await?;
// マージンがデータからはみ出さないことを保証
// cf. https://github.com/VOICEVOX/voicevox_core/pull/854#discussion_r1803691291
@@ -1024,20 +1024,19 @@ mod inner {
}
/// 与えられた音声特徴量で音声生成。
- /// CPU/GPU-boundな操作なので、非同期ランタイム上では直接実行されるべきではない。
- fn render_audio_segment(
+ pub(super) async fn render_audio_segment<A: infer::AsyncExt>(
&self,
spec: ndarray::Array2<f32>,
style_id: StyleId,
) -> Result<ndarray::Array1<f32>> {
let (model_id, _inner_voice_id) = self.ids_for::<TalkDomain>(style_id)?;
- let RenderAudioSegmentOutput { wave } =
- self.run_session(model_id, RenderAudioSegmentInput { spec })?;
+ let RenderAudioSegmentOutput { wave } = self
+ .run_session::<A, _>(model_id, RenderAudioSegmentInput { spec })
+ .await?;
Ok(wave)
}
- /// CPU/GPU-boundな操作なので、非同期ランタイム上では直接実行されるべきではない。
- fn decode(
+ pub(super) async fn decode<A: infer::AsyncExt>(
&self,
length: usize,
phoneme_size: usize,
@@ -1045,14 +1044,12 @@ mod inner {
phoneme_vector: ndarray::Array1<f32>,
style_id: StyleId,
) -> Result<Vec<f32>> {
- let intermediate = self.generate_full_intermediate(
- length,
- phoneme_size,
- f0,
- phoneme_vector,
- style_id,
- )?;
- let output_with_margin = self.render_audio_segment(intermediate, style_id)?;
+ let intermediate = self
+ .generate_full_intermediate::<A>(length, phoneme_size, f0, phoneme_vector, style_id)
+ .await?;
+ let output_with_margin = self
+ .render_audio_segment::<A>(intermediate, style_id)
+ .await?;
let output = trim_margin_from_wave(output_with_margin);
Ok(output.to_vec())
}
diff --git a/crates/voicevox_core_macros/src/inference_domain.rs b/crates/voicevox_core_macros/src/inference_domain.rs
index f959982e4..6570791d2 100644
--- a/crates/voicevox_core_macros/src/inference_domain.rs
+++ b/crates/voicevox_core_macros/src/inference_domain.rs
@@ -222,17 +222,22 @@ pub(crate) fn derive_inference_input_signature(
fn make_run_context<R: crate::infer::InferenceRuntime>(
self,
- sess: &mut R::Session,
- ) -> ::anyhow::Result<R::RunContext<'_>> {
- let mut ctx = <R::RunContext<'_> as ::std::convert::From<_>>::from(sess);
+ sess: ::std::sync::Arc<R::Session>,
+ ) -> ::anyhow::Result<R::RunContext> {
+ let mut ctx = <R::RunContext as ::std::convert::From<_>>::from(sess);
#(
- __ArrayExt::push_to_ctx(self.#field_names, &mut ctx)?;
+ __ArrayExt::push_to_ctx(
+ self.#field_names,
+ ::std::stringify!(#field_names),
+ &mut ctx,
+ )?;
)*
return ::std::result::Result::Ok(ctx);
trait __ArrayExt {
fn push_to_ctx(
self,
+ name: &'static str,
ctx: &mut impl crate::infer::PushInputTensor,
) -> ::anyhow::Result<()>;
}
@@ -242,9 +247,10 @@ pub(crate) fn derive_inference_input_signature(
{
fn push_to_ctx(
self,
+ name: &'static str,
ctx: &mut impl crate::infer::PushInputTensor,
) -> ::anyhow::Result<()> {
- A::push_tensor_to_ctx(self, ctx)
+ A::push_tensor_to_ctx(name, self, ctx)
}
}
}
diff --git a/crates/voicevox_core_python_api/python/test/test_asyncio_user_dict_load.py b/crates/voicevox_core_python_api/python/test/test_asyncio_user_dict_load.py
index e433a7442..135822ca7 100644
--- a/crates/voicevox_core_python_api/python/test/test_asyncio_user_dict_load.py
+++ b/crates/voicevox_core_python_api/python/test/test_asyncio_user_dict_load.py
@@ -6,6 +6,7 @@
# AudioQueryのkanaを比較して変化するかどうかで判断する。
+import multiprocessing
from uuid import UUID
import conftest # noqa: F401
@@ -20,7 +21,13 @@ async def test_user_dict_load() -> None:
)
open_jtalk = await voicevox_core.asyncio.OpenJtalk.new(conftest.open_jtalk_dic_dir)
model = await voicevox_core.asyncio.VoiceModelFile.open(conftest.model_dir)
- synthesizer = voicevox_core.asyncio.Synthesizer(onnxruntime, open_jtalk)
+ synthesizer = voicevox_core.asyncio.Synthesizer(
+ onnxruntime,
+ open_jtalk,
+ cpu_num_threads=max(
+ multiprocessing.cpu_count(), 2
+ ), # https://github.com/VOICEVOX/voicevox_core/issues/888
+ )
await synthesizer.load_voice_model(model)
diff --git a/example/python/run-asyncio.py b/example/python/run-asyncio.py
index 0d7460f90..3cb8a7e48 100644
--- a/example/python/run-asyncio.py
+++ b/example/python/run-asyncio.py
@@ -4,6 +4,7 @@
import dataclasses
import json
import logging
+import multiprocessing
from argparse import ArgumentParser
from pathlib import Path
@@ -91,7 +92,12 @@ async def main() -> None:
logger.info("%s", f"Initializing ({args.mode=}, {args.dict_dir=})")
synthesizer = Synthesizer(
- onnxruntime, await OpenJtalk.new(args.dict_dir), acceleration_mode=args.mode
+ onnxruntime,
+ await OpenJtalk.new(args.dict_dir),
+ acceleration_mode=args.mode,
+ cpu_num_threads=max(
+ multiprocessing.cpu_count(), 2
+ ), # https://github.com/VOICEVOX/voicevox_core/issues/888
)
logger.debug("%s", f"{synthesizer.metas=}")
diff --git a/example/python/run.py b/example/python/run.py
index 4a22e709c..5970f5dff 100644
--- a/example/python/run.py
+++ b/example/python/run.py
@@ -1,6 +1,7 @@
import dataclasses
import json
import logging
+import multiprocessing
from argparse import ArgumentParser
from pathlib import Path
@@ -95,7 +96,12 @@ def main() -> None:
logger.info("%s", f"Initializing ({args.mode=}, {args.dict_dir=})")
synthesizer = Synthesizer(
- onnxruntime, OpenJtalk(args.dict_dir), acceleration_mode=args.mode
+ onnxruntime,
+ OpenJtalk(args.dict_dir),
+ acceleration_mode=args.mode,
+ cpu_num_threads=max(
+ multiprocessing.cpu_count(), 2
+ ), # https://github.com/VOICEVOX/voicevox_core/issues/888
)
logger.debug("%s", f"{synthesizer.metas=}")
From ac6ca3f5a4936d89715a423d2c72f247ab2eca0f Mon Sep 17 00:00:00 2001
From: Ryo Yamashita <qryxip@gmail.com>
Date: Mon, 9 Dec 2024 14:10:26 +0900
Subject: [PATCH 2/2] Minor refactor
---
crates/voicevox_core/src/infer.rs | 18 ++++++++----------
.../src/infer/runtimes/onnxruntime.rs | 2 +-
2 files changed, 9 insertions(+), 11 deletions(-)
diff --git a/crates/voicevox_core/src/infer.rs b/crates/voicevox_core/src/infer.rs
index 16b246e36..7b81b7c5d 100644
--- a/crates/voicevox_core/src/infer.rs
+++ b/crates/voicevox_core/src/infer.rs
@@ -3,7 +3,7 @@ mod model_file;
pub(crate) mod runtimes;
pub(crate) mod session_set;
-use std::{borrow::Cow, collections::BTreeSet, fmt::Debug, future::Future, ops::Index, sync::Arc};
+use std::{borrow::Cow, collections::BTreeSet, fmt::Debug, ops::Index, sync::Arc};
use derive_new::new;
use duplicate::duplicate_item;
@@ -27,7 +27,7 @@ impl AsyncExt for SingleTasked {
async fn run_session<R: InferenceRuntime>(
ctx: R::RunContext,
) -> anyhow::Result<Vec<OutputTensor>> {
- R::run(ctx)
+ R::run_blocking(ctx)
}
}
@@ -43,7 +43,7 @@ pub(crate) trait InferenceRuntime: 'static {
// TODO: "session"とは何なのかを定め、ドキュメントを書く。`InferenceSessionSet`も同様。
type Session;
- // 本当は`From<'_ Self::Session>`としたいが、 rust-lang/rust#100013 がある
+ // 本当は`From<&'_ Self::Session>`としたいが、 rust-lang/rust#100013 が立ち塞がる
type RunContext: From<Arc<Self::Session>> + PushInputTensor;
/// 名前。
@@ -70,11 +70,9 @@ pub(crate) trait InferenceRuntime: 'static {
Vec<ParamInfo<OutputScalarKind>>,
)>;
- fn run(ctx: Self::RunContext) -> anyhow::Result<Vec<OutputTensor>>;
+ fn run_blocking(ctx: Self::RunContext) -> anyhow::Result<Vec<OutputTensor>>;
- fn run_async(
- ctx: Self::RunContext,
- ) -> impl Future<Output = anyhow::Result<Vec<OutputTensor>>> + Send;
+ async fn run_async(ctx: Self::RunContext) -> anyhow::Result<Vec<OutputTensor>>;
}
/// 共に扱われるべき推論操作の集合を示す。
@@ -115,7 +113,7 @@ pub(crate) trait InferenceOperation: Copy + Enum {
/// `InferenceDomain`の推論操作を表す列挙型。
///
/// `::macros::InferenceOperation`により、具体型ごと生成される。
-pub(crate) trait InferenceSignature: Sized + Send + 'static {
+pub(crate) trait InferenceSignature {
type Domain: InferenceDomain;
type Input: InferenceInputSignature<Signature = Self>;
type Output: InferenceOutputSignature;
@@ -125,7 +123,7 @@ pub(crate) trait InferenceSignature: Sized + Send + 'static {
/// 推論操作の入力シグネチャ。
///
/// `::macros::InferenceInputSignature`により導出される。
-pub(crate) trait InferenceInputSignature: Send + 'static {
+pub(crate) trait InferenceInputSignature {
type Signature: InferenceSignature<Input = Self>;
const PARAM_INFOS: &'static [ParamInfo<InputScalarKind>];
fn make_run_context<R: InferenceRuntime>(
@@ -189,7 +187,7 @@ pub(crate) trait PushInputTensor {
///
/// `::macros::InferenceOutputSignature`により、`TryFrom<OutputTensor>`も含めて導出される。
pub(crate) trait InferenceOutputSignature:
- TryFrom<Vec<OutputTensor>, Error = anyhow::Error> + Send
+ TryFrom<Vec<OutputTensor>, Error = anyhow::Error>
{
const PARAM_INFOS: &'static [ParamInfo<OutputScalarKind>];
}
diff --git a/crates/voicevox_core/src/infer/runtimes/onnxruntime.rs b/crates/voicevox_core/src/infer/runtimes/onnxruntime.rs
index 8e457bbf0..f2cc4fac7 100644
--- a/crates/voicevox_core/src/infer/runtimes/onnxruntime.rs
+++ b/crates/voicevox_core/src/infer/runtimes/onnxruntime.rs
@@ -182,7 +182,7 @@ impl InferenceRuntime for self::blocking::Onnxruntime {
Ok((sess.into(), input_param_infos, output_param_infos))
}
- fn run(
+ fn run_blocking(
OnnxruntimeRunContext { sess, inputs }: Self::RunContext,
) -> anyhow::Result<Vec<OutputTensor>> {
extract_outputs(&sess.lock_blocking().run(inputs)?)