Skip to content

Fix AMD rocm #92

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Alexander-Ger-Reich wants to merge 3 commits into
base: master
Choose a base branch
from
Open

Fix AMD rocm #92

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
dfa0aef
AMD rocm Fix 1
Alexander-Ger-Reich Apr 12, 2026
0261153
AMD rocm Fix 2
Alexander-Ger-Reich Apr 12, 2026
3382828
AMD rocm Fix 3
Alexander-Ger-Reich Apr 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions omnivoice/utils/audio.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,7 @@
import numpy as np
import torch
import torchaudio
import soundfile as sf
from pydub import AudioSegment
from pydub.silence import detect_leading_silence, detect_nonsilent, split_on_silence

Expand All @@ -42,9 +43,8 @@ def load_audio(audio_path: str, sampling_rate: int):
PyTorch tensor of shape (1, T)
"""
try:
waveform, prompt_sampling_rate = torchaudio.load(
audio_path, backend="soundfile"
)
data, prompt_sampling_rate = sf.read(audio_path, dtype="float32")
waveform = torch.from_numpy(data.T if data.ndim > 1 else data.reshape(1, -1))
except (RuntimeError, OSError):
# Fallback via pydub+ffmpeg for formats torchaudio can't handle
aseg = AudioSegment.from_file(audio_path)
Expand Down
26 changes: 21 additions & 5 deletions pyproject.toml
View file
Open in desktop

@EnochPrime EnochPrime May 5, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can avoid needed users to uncomment lines by using the optional dependencies methods like in #143.

Original file line number Diff line number Diff line change
Expand Up @@ -68,13 +68,29 @@ Repository = "https://github.com/k2-fsa/OmniVoice"
# Install PyTorch with CUDA support on Linux/Windows (CUDA doesn't exist for Mac).
# NOTE: We must explicitly request them as `dependencies` above. These improved
# versions will not be selected if they're only third-party dependencies.

# =========================================================
# IMPORTANT: Select the correct graphics card here!
# Comment out the block for your current card
# and uncomment the other one.
# =========================================================

# FOR NVIDIA:
torch = [
{ index = "pytorch-cuda", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
]
torchaudio = [
{ index = "pytorch-cuda", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
]

# FOR AMD:
#torch = [
# { index = "pytorch-rocm", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
#]
#torchaudio = [
# { index = "pytorch-rocm", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
#]

[[tool.uv.index]]
name = "pytorch-cuda"
# Use PyTorch built for NVIDIA Toolkit version 12.8.
Expand All @@ -83,11 +99,11 @@ url = "https://download.pytorch.org/whl/cu128"
# Only use this index when explicitly requested by `tool.uv.sources`.
explicit = true

[tool.uv]
constraint-dependencies = [
"torch==2.8.0",
"torchaudio==2.8.0",
]
[[tool.uv.index]]
name = "pytorch-rocm"
# For AMD (e.g., RX 7900 series / GFX 1100 architecture)
url = "https://rocm.nightlies.amd.com/v2/gfx110X-all/"
explicit = true

[tool.hatch.build.targets.sdist]
include = ["omnivoice"]
Expand Down