If GPU operations can be implemented, the extraction speed should be much faster

[martjay]

If GPU operations can be implemented, the extraction speed should be much faster, The CPU mode is still too slow.

[martjay]

https://github.com/openai/whisper/discussions/47

[martjay]

https://s3.bmp.ovh/imgs/2023/01/04/936aad75fd9795aa.jpg

I can't find .srt file

[chidiwilliams]

Sure, I haven't figured out the GPU yet. But I'll mark this as a duplicate of #182 for now.

[chidiwilliams]

https://s3.bmp.ovh/imgs/2023/01/04/936aad75fd9795aa.jpg

I can't find .srt file

I don't have access to the image. Maybe a permissions issue?

[martjay]

https://s3.bmp.ovh/imgs/2023/01/04/936aad75fd9795aa.jpg I can't find .srt file

I don't have access to the image. Maybe a permissions issue?

[xxx975]

u can try select this line and click third ico

[xxx975]

this line == 3 Levels of Chords - ........
:)

[martjay]

this line == 3 Levels of Chords - ........ :)

Thank you. In this way, the subtitles can be exported. I think the subtitles will be automatically exported to the video directory

[esonec]

Followed the steps from here https://github.com/openai/whisper/discussions/47

For me now: "CUDA enabled: True"

But the Buzz still runs on the CPU. While they say, Whisper should detect CUDA and run on GPU.

[chidiwilliams]

this line == 3 Levels of Chords - ........ :)

Thank you. In this way, the subtitles can be exported. I think the subtitles will be automatically exported to the video directory

Yes, the behaviour changed in the last release. Double-clicking the line also works :)

[martjay]

Followed the steps from here openai/whisper#47

For me now: "CUDA enabled: True"

But the Buzz still runs on the CPU. While they say, Whisper should detect CUDA and run on GPU.

Using GPU mode will be much faster than using CPU. I really don't cheat you

[martjay]

this line == 3 Levels of Chords - ........ :)

Thank you. In this way, the subtitles can be exported. I think the subtitles will be automatically exported to the video directory

Yes, the behaviour changed in the last release. Double-clicking the line also works :)

Can't the latest version automatically generate the srt to the video directory? I think automatic generation is more convenient

[ttimasdf]

Currently, pytorch with CUDA support is only available if you manually install pytorch build supporting CUDA inside poetry virtualenv. You also need to uninstall previous build with pip.

https://github.com/openai/whisper/discussions/47#discussioncomment-4549820

Follow the steps above, then apply the following patch.

diff --git a/buzz/transcriber.py b/buzz/transcriber.py
index c087c13..c489d9a 100644
--- a/buzz/transcriber.py
+++ b/buzz/transcriber.py
@@ -424,7 +424,8 @@ def transcribe_whisper(stderr_conn: Connection, task: FileTranscriptionTask):
                                       task=task.transcription_options.task.value, verbose=False)
             whisper_segments = result.get('segments')
         else:
-            model = whisper.load_model(task.model_path)
+            logging.info("Loading whisper with CUDA device")
+            model = whisper.load_model(task.model_path, device="cuda")
             if task.transcription_options.word_level_timings:
                 stable_whisper.modify_model(model)
                 result = model.transcribe(

[ttimasdf]

Changes I made to install pytorch through poetry. Tested on windows.

diff --git a/pyproject.toml b/pyproject.toml
index 2ffef1f..12a60ea 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -10,13 +10,15 @@ readme = "README.md"
 python = ">=3.9.13,<3.11"
@@ -10,13 +10,15 @@ readme = "README.md"
 python = ">=3.9.13,<3.11"
 sounddevice = "^0.4.5"
 whisper = { git = "https://github.com/openai/whisper.git" }
-torch = "1.12.1"
+torch = {version = "^1.13.1+cu117", source = "pytorch-cu117"}
 transformers = "~4.24.0"
 appdirs = "^1.4.4"
 ffmpeg-python = "^0.2.0"
 humanize = "^4.4.0"
 PyQt6 = "^6.4.0"
+stable-ts = "^1.0.2"
 tqdm = "^4.64.1"
+click = "^8.1.3"

 [tool.poetry.group.dev.dependencies]
 autopep8 = "^1.7.0"
@@ -30,6 +32,13 @@ pylint = "^2.15.5"
 pre-commit = "^2.20.0"
 ctypesgen = "^1.1.1"

+
+[[tool.poetry.source]]
+name = "pytorch-cu117"
+url = "https://download.pytorch.org/whl/cu117"
+default = false
+secondary = true
+
 [build-system]
 requires = ["poetry-core"]
 build-backend = "poetry.core.masonry.api"

[martjay]

Changes I made to install pytorch through poetry. Tested on windows.

diff --git a/pyproject.toml b/pyproject.toml
index 2ffef1f..12a60ea 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -10,13 +10,15 @@ readme = "README.md"
 python = ">=3.9.13,<3.11"
@@ -10,13 +10,15 @@ readme = "README.md"
 python = ">=3.9.13,<3.11"
 sounddevice = "^0.4.5"
 whisper = { git = "https://github.com/openai/whisper.git" }
-torch = "1.12.1"
+torch = {version = "^1.13.1+cu117", source = "pytorch-cu117"}
 transformers = "~4.24.0"
 appdirs = "^1.4.4"
 ffmpeg-python = "^0.2.0"
 humanize = "^4.4.0"
 PyQt6 = "^6.4.0"
+stable-ts = "^1.0.2"
 tqdm = "^4.64.1"
+click = "^8.1.3"

 [tool.poetry.group.dev.dependencies]
 autopep8 = "^1.7.0"
@@ -30,6 +32,13 @@ pylint = "^2.15.5"
 pre-commit = "^2.20.0"
 ctypesgen = "^1.1.1"

+
+[[tool.poetry.source]]
+name = "pytorch-cu117"
+url = "https://download.pytorch.org/whl/cu117"
+default = false
+secondary = true
+
 [build-system]
 requires = ["poetry-core"]
 build-backend = "poetry.core.masonry.api"

@chidiwilliams see, but I dont know how to do, pls update new version release~

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.