diff --git a/sci-libs/torchaudio/Manifest b/sci-libs/torchaudio/Manifest
new file mode 100644
index 00000000000..0059647a605
--- /dev/null
+++ b/sci-libs/torchaudio/Manifest
@@ -0,0 +1 @@
+DIST torchaudio-2.4.1.tar.gz 4503000 BLAKE2B 85ae91bb564996f95c9116802058968dff191d4771becbb859987afd3b78189980b31281c5c49088fe91a72a02170dafe865f8bbb4168af6aeafe42f4029c09b SHA512 84c196a19bedefd39c7c1bc690705e317aaba9e16aa52a71dfa5688faf50ffa1985008c59007939ad7096c7900c5450659efb12e0a393c4545fe4bd31f294b11
diff --git a/sci-libs/torchaudio/metadata.xml b/sci-libs/torchaudio/metadata.xml
new file mode 100644
index 00000000000..445ee5c8c44
--- /dev/null
+++ b/sci-libs/torchaudio/metadata.xml
@@ -0,0 +1,11 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE pkgmetadata SYSTEM "http://www.gentoo.org/dtd/metadata.dtd">
+<pkgmetadata>
+	<use>
+		<flag name="rocm">Enable ROCm gpu computing support</flag>
+		<flag name="rnnt">Enable custom RNN-T loss function</flag>
+	</use>
+	<upstream>
+		<remote-id type="github">pytorch/vision</remote-id>
+	</upstream>
+</pkgmetadata>
diff --git a/sci-libs/torchaudio/torchaudio-2.4.1.ebuild b/sci-libs/torchaudio/torchaudio-2.4.1.ebuild
new file mode 100644
index 00000000000..dae78df2471
--- /dev/null
+++ b/sci-libs/torchaudio/torchaudio-2.4.1.ebuild
@@ -0,0 +1,71 @@
+# Copyright 2020-2024 Gentoo Authors
+# Distributed under the terms of the GNU General Public License v2
+
+EAPI=8
+
+PYTHON_COMPAT=( python3_{11..12} )
+DISTUTILS_SINGLE_IMPL=1
+DISTUTILS_USE_PEP517=setuptools
+DISTUTILS_EXT=1
+ROCM_SKIP_GLOBALS=1
+inherit cuda distutils-r1 multiprocessing rocm optfeature
+
+DESCRIPTION="Data manipulation and transformation for audio signal processing"
+HOMEPAGE="https://github.com/pytorch/audio"
+SRC_URI="https://github.com/pytorch/audio/archive/v${PV}.tar.gz -> ${P}.tar.gz"
+S="${WORKDIR}/audio-${PV}"
+
+LICENSE="BSD"
+SLOT="0"
+KEYWORDS="~amd64"
+IUSE="cuda rocm ffmpeg openmp +rnnt"
+REQUIRED_USE="?? ( cuda rocm )"
+
+DEPEND="
+	ffmpeg? ( media-video/ffmpeg:= )
+	sci-libs/caffe2[cuda?,rocm?,openmp?]
+	sci-libs/pytorch[${PYTHON_SINGLE_USEDEP}]
+"
+RDEPEND="
+	${DEPEND}
+"
+BDEPEND="
+	test? (
+		$(python_gen_cond_dep '
+			dev-python/expecttest[${PYTHON_USEDEP}]
+			dev-python/numpy[${PYTHON_USEDEP}]
+			dev-python/typing-extensions[${PYTHON_USEDEP}]
+			dev-python/parameterized[${PYTHON_USEDEP}]
+			dev-python/scipy[${PYTHON_USEDEP}]
+			dev-python/scikit-learn[${PYTHON_USEDEP}]
+			dev-python/soundfile[${PYTHON_USEDEP}]
+		')
+	)
+"
+
+distutils_enable_tests pytest
+
+src_compile() {
+	export MAX_JOBS="$(makeopts_jobs)" # Let ninja respect MAKEOPTS
+
+	export USE_CUDA=$(usex cuda 1 0)
+	export USE_ROCM=$(usex rocm 1 0)
+	use rocm && addpredict /dev/kfd
+	export USE_OPENMP=$(usex openmp 1 0)
+	export USE_FFMPEG=$(usex ffmpeg 1 0)
+	export BUILD_SOX=0
+	export BUILD_RNNT=$(usex rnnt 1 0)
+	use ffmpeg && export FFMPEG_ROOT=${EPREFIX}/usr
+
+	distutils-r1_src_compile
+}
+
+python_test() {
+	use rocm && check_amdgpu
+
+	epytest -p expecttest
+}
+
+pkg_postinst() {
+	optfeature "SoundFile I/O backend" dev-python/soundfile
+}