Hello could you let me know how to install and run your program on windows. I am on the step where I did "pip3 install -r requirements.txt" and I got the following error.

        ERROR: Cannot install -r requirements.txt (line 4) and numpy==1.18.1 because these package versions have conflicting 
        dependencies.
        
        The conflict is caused by:
            The user requested numpy==1.18.1
            deepspeech 0.8.2 depends on numpy<=1.17.0 and >=1.14.5
        
        To fix this you could try to:
        1. loosen the range of package versions you've specified
        2. remove package versions to allow pip attempt to solve the dependency conflict
        
        ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/user_guide/#fixing-conflicting-dependencies

After the mozilla layoffs, the deepspeech team forked the Deepspeech repo and founded the company Coqui AI (https://github.com/coqui-ai/STT) where they continue the development and AFAIK they now only allow .tflite files to export models. It theoretically should work with the old code, but for me it didn't.

When I try to run it like this:

python3 autosub/main.py --file /Users/sgrotz/Downloads/kp193-hejma-auxtomatigo.mp3 --split-duration 8

with a .tflite file in the main folder and NO language model.

Then I get:

AutoSub

['autosub/main.py', '--file', '/Users/sgrotz/Downloads/kp193-hejma-auxtomatigo.mp3', '--split-duration', '8']
ARGS: Namespace(dry_run=False, file='/Users/sgrotz/Downloads/kp193-hejma-auxtomatigo.mp3', format=['srt', 'vtt', 'txt'], model=None, scorer=None, split_duration=8.0)
Warning no models specified via --model and none found in local directory. Please run getmodel.sh convenience script from autosub repo to get some.
Error: Must have pbmm model. Exiting

Have I done anything wrong here or doesn't AutoSub support .rflite files?

I tested it on MacOS and installed ffmpeg via homebrew.

Added GPU support for Docker images and a bunch of other related changes to ensure best possible docker experience.

Major Changes: Base Image is now Ubuntu so we can use NVIDIA's cuda images Dockerfile was significantly revamped Dropped unnecessary bit-rate conversion from ffmpeg command Reordered code to allow instantiating tensor model before failure due to inadequate arguments allowing for useful side effect of faster startup times Added simple build scripts There were many errors regarding proper parsing of Paths with metacharacters. They have been fixed where I found them. We should probably stick to using os.path functions instead of manually parsing the strings ourselves.

Also: Audio directory is directly emptied rather than deleted and recreated. Makes a difference for running docker in Read only mode. Option to ask for overwriting existing SRT. Default is not to overwrite. Previous default was append, which is not useful. Dockerfile now just copies training model if available instead of downloading each time. This enables better developement experience so I don't have to keep downloading each time I was experimenting with getting a GPU build. Users can still use newly created convenience script called getmodels.sh to still automatically get what they want. Other minor improvements like ensuring apt is non-interactive and clears its cache after installing. a specific series of copy commands to only copy what is required rather than entire project dir which may have other things users choose to use.

pip3 install -r requirements.txt

Collecting deepspeech==0.9.3 (from -r requirements.txt (line 3))
  ERROR: Could not find a version that satisfies the requirement deepspeech==0.9.3 (from -r requirements.txt (line 3)) (from versions: none)
ERROR: No matching distribution found for deepspeech==0.9.3 (from -r requirements.txt (line 3))

Hi, after I installed Autosub I always get this error message when I try to run the program "ImportError: DLL load failed while importing _impl: A dynamic link library (DLL) initialization routine failed." Could anyone tell me how to fix this?

Hi, my config is Win10_x64, Python 3.8. When i execute $ C:/Soft/Autosub/sub/Scripts/python autosub/main.py --file D:/Work/video.mkv. It gives me error: Traceback (most recent call last): File "autosub/main.py", line 8, in <module> from . import logger Info: `[email protected] MINGW64 /c/Soft/Autosub (master) $ pip list Package Version

absl-py 1.0.0 astunparse 1.6.3 cachetools 4.2.4 certifi 2021.10.8 charset-normalizer 2.0.12 cycler 0.10.0 deepspeech-gpu 0.9.3 distlib 0.3.4 ffmpeg 1.4 filelock 3.6.0 gast 0.3.3 google-auth 1.35.0 google-auth-oauthlib 0.4.6 google-pasta 0.2.0 grpcio 1.44.0 h5py 2.10.0 idna 3.3 importlib-metadata 4.11.3 joblib 0.16.0 Keras-Preprocessing 1.1.2 kiwisolver 1.2.0 Markdown 3.3.6 numpy 1.22.3 oauthlib 3.2.0 opt-einsum 3.3.0 pip 19.2.3 platformdirs 2.5.1 protobuf 3.19.4 pyasn1 0.4.8 pyasn1-modules 0.2.8 pydub 0.23.1 pyparsing 2.4.7 python-dateutil 2.8.1 requests 2.27.1 requests-oauthlib 1.3.1 rsa 4.8 scikit-learn 1.0.2 scipy 1.4.1 setuptools 41.2.0 six 1.15.0 stt 1.0.0 tensorboard 2.2.2 tensorboard-plugin-wit 1.8.1 tensorflow-gpu 2.2.0 tensorflow-gpu-estimator 2.2.0 termcolor 1.1.0 threadpoolctl 3.1.0 tqdm 4.44.1 urllib3 1.26.9 virtualenv 20.13.3 Werkzeug 2.0.3 wheel 0.37.1 wrapt 1.14.0 zipp 3.7.0 WARNING: You are using pip version 19.2.3, however version 22.0.4 is available. You should consider upgrading via the 'python -m pip install --upgrade pip' command.`

Fixes #21 Fixes #31 The 3 digit millisecond commit cherry-picked of public fork. It seemed pretty reasonable to include it here.

See commit messages and referenced issues for more information.

I will cherry-pick the best changes from all the public forks in the coming days and weeks so that AutoSub can become what it should be, and I intend to help you actively maintain and develop this project over the long term.

Thanks!

I was just exploring the code but wanted to tidy up a bit while I was at it.

EditorConfig

Adds an EditorConfig file, so contributors can use consistent code styles with the rest of the project. This makes it more convenient to work on the project when a developer's editor settings aren't the same as yours.

Optimizes the Dockerfile

This reduces the number of intermediate layers, and makes the build marginally faster. Its good practice to minimize layers in a Docker image to make the final image smaller, especially if there's a lot of file system changes between layers. In this case, the difference in size is negligible.

A Docker image consists of read-only layers each of which represents a Dockerfile instruction. The layers are stacked and each one is a delta of the changes from the previous layer. … Each instruction creates one layer:

FROM creates a layer from the ubuntu:18.04 Docker image. COPY adds files from your Docker client’s current directory. RUN builds your application with make. CMD specifies what command to run within the container.

— https://docs.docker.com/develop/develop-images/dockerfile_best-practices/

Optimize apt-get

• Removes the cached lists from /var/lib/apt/lists. (Reduces image size by 0.02 GB.)
• No longer installs recommended packages. (Reduces image size by another 0.02 GB.)

More info:

• https://ubuntu.com/blog/we-reduced-our-docker-images-by-60-with-no-install-recommends
• https://hackernoon.com/tips-to-reduce-docker-image-sizes-876095da3b34

Before

$ time docker build -t autosub:dev --build-arg model=0.9.3 --no-cache .
real    3m57.379s
user    0m1.650s
sys     0m1.776s

$ docker images -a
REPOSITORY                  TAG               IMAGE ID       CREATED              SIZE
autosub                     dev               2565f52660cb   About a minute ago   1.86GB
<none>                      <none>            4fd064c626ea   About a minute ago   1.86GB
<none>                      <none>            d9d3ae1ed1db   About a minute ago   1.86GB
<none>                      <none>            cf1540452ffc   2 minutes ago        714MB
<none>                      <none>            dfa2fae14669   2 minutes ago        714MB
<none>                      <none>            9ada5732266d   2 minutes ago        714MB
<none>                      <none>            d6821a416a2c   3 minutes ago        416MB
<none>                      <none>            c918464fff64   3 minutes ago        416MB
<none>                      <none>            594b7f697317   5 minutes ago        114MB
<none>                      <none>            e5f83a59a602   5 minutes ago        114MB
python                      3.8-slim-buster   4728acd2148c   2 days ago           114MB

After

$ time docker build -t autosub:dev --build-arg model=0.9.3 --no-cache .
real    3m38.463s
user    0m1.252s
sys     0m1.209s

$ docker images -a
REPOSITORY                  TAG               IMAGE ID       CREATED          SIZE
autosub                     dev               41dbaca6c36b   7 minutes ago    1.82GB
<none>                      <none>            117e3c54c5cf   7 minutes ago    1.82GB
<none>                      <none>            a911683cc89d   10 minutes ago   114MB
<none>                      <none>            7cf695bce4db   10 minutes ago   114MB
<none>                      <none>            264a99e3d9d1   10 minutes ago   114MB
python                      3.8-slim-buster   4728acd2148c   2 days ago       114MB

docker build -t autosub . Cannot build and results in

Step 11/13 : RUN pip3 install --no-cache-dir -r requirements.txt
 ---> Running in bdf3fc44f538
Collecting cycler==0.10.0 (from -r requirements.txt (line 1))
  Downloading https://files.pythonhosted.org/packages/f7/d2/e07d3ebb2bd7af696440ce7e754c59dd546ffe1bbe732c8ab68b9c834e61/cycler-0.10.0-py2.py3-none-any.whl
Collecting numpy (from -r requirements.txt (line 2))
  Downloading https://files.pythonhosted.org/packages/45/b2/6c7545bb7a38754d63048c7696804a0d947328125d81bf12beaa692c3ae3/numpy-1.19.5-cp36-cp36m-manylinux1_x86_64.whl (13.4MB)
Collecting stt==1.0.0 (from -r requirements.txt (line 3))
  Could not find a version that satisfies the requirement stt==1.0.0 (from -r requirements.txt (line 3)) (from versions: 0.10.0a5, 0.10.0a6, 0.10.0a8, 0.10.0a9, 0.10.0a10)
No matching distribution found for stt==1.0.0 (from -r requirements.txt (line 3))
The command '/bin/sh -c pip3 install --no-cache-dir -r requirements.txt' returned a non-zero code: 1

Is it impossible to recognize in another language?

I hope the caption file comes out in Japanese.

I put a video of the conversation in Japanese and it came out in English, is there a way to modify it?

or.. is there any japaense model files?

Hi,

I did this manually but maybe someone can improve it and write some script for it?:

1. Check text that is longer than 7 words

2. Add a line break to each line longer than 7 words.

3. Get the numbers of lines you got, ie: blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah

There are 4 lines 4. We take the initial and final time

13 <<< SRT Subtitle position ie: 00:00:25,90 --> 00:00:35,25

5. We surplus them and divide them by the number of lines 35,25 - 25,90 = 9,35 4 lines of max 7 words each 9,35/4 = 2,33

6. We take 2.33 less 1 for each limit, Ie:

13 <<< SRT Subtitle position 00:00:25,90 --> 00:00:28,23 blah blah blah blah blah blah blah

14 00:00:28,24 --> 00:00:30,57 blah blah blah blah blah blah blah

15 00:00:30,58 --> 00:00:32,91 blah blah blah blah blah blah blah

16 00:00:32,92 --> 00:00:35,25 blah blah blah blah blah

14 <<<< WARNING >>>> update all the other numbers in this case to 17 (check step 7 below) blah blah blah blah

7. We need to update the SRT Subtitle position, in this case we finished at 16, so we need to replace 14 for 17, and do the same for all the other numbers. Note: Update from top to bottom so the counter increments.

That's it.

Anyone? :)

Otherwise encounter error

    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "C:\Users\Aqaao\AppData\Local\Temp\pip-req-build-2dcr43hl\setup.py", line 8, in <module>
        README = fh.read()
    UnicodeDecodeError: 'gbk' codec can't decode byte 0x90 in position 757: illegal multibyte sequence

Adds support for Python 3.10.

One of the major problems upgrading to Python 3.10 with this is that DeepSpeech appears to be unmaintained and no longer supports 3.10. So this also removes DeepSpeech as an engine but also updates the getmodels.sh script to easily download Coqui models.

I build the docker image as the instruction. It returns this error after running.

Traceback (most recent call last):
  File "./autosub/main.py", line 10, in <module>
    from autosub import logger
ModuleNotFoundError: No module named 'autosub'

I think the problem is I didn't have autosub installed in the image. So I add RUN pip3 install . to Dockerfile. Then everything works out fine.

BTW: With this solution, it's also necessary to have COPY README.md ./ in the image. Also, I have to add encoding='utf-8' in with open("README.md", "r") as fh: in setup.py, otherwise it would use ASCII encoding as default in my case.

When I first started using this program, I took subtitles and opened them in VLC and it looked fine. But then I tried another program, and they refused to work. Took me quite a while to realize that the output of AutoSub was actually not compliant with the VTT format according to this validator:

https://w3c.github.io/webvtt.js/parser.html

Errors I get include:

You are hopeless, RTFS. (10ms)
1. Line 2: No blank line after the signature.
2. Line 7: Cue identifier cannot be standalone.
3. Line 11: Cue identifier cannot be standalone.
4. Line 15: Cue identifier cannot be standalone.
5. Line 19: Cue identifier cannot be standalone.
6. Line 23: Cue identifier cannot be standalone.
7. Line 27: Cue identifier cannot be standalone.
[many, many more of these]

Unfortunately, I can't share the subtitles that did this. However, it happened with every file I tried. I was able to build a series of sed replacements to run on the AutoSub output file to make it so it passes the validator, but that's quite a hack. I'd recommend trying validation on a file yourself - it was consistently repeatable for me.

OpenAI just released probably the best model that there is for speech recognition right now.

It would be great to incorprate this into this project!

More info: https://openai.com/blog/whisper/

autosub is using only 100% cpu when it should use 400% on a quad core cpu

the task should be easy to do in parallel, by splitting the audio into N segments for N cpu cores

related

• https://stackoverflow.com/questions/4047789/parallel-file-parsing-multiple-cpu-cores
  • multiprocessing.Pool
  • ray for distributed computing