Windows 10 uses DeepSpeech: v0.7.0-0-g3fbbca2b even after upgrading to v0.7.4

[SanghviChirag]

Hello,
I've been using deepspeech model in one of my projects. Thanks to this, my work has become easy.

I'm developing desktop apps for Windows, Mac and Linux, and in all of them, I'm using the latest stable version i.e. 0.7.4.
In Windows, even after using Deepspeech v0.7.4, it shows that it is using the previous version for transcription.
TensorFlow: v1.15.0-24-gceb46aae58
DeepSpeech: v0.7.0-0-g3fbbca2b

while for Mac and Linux, it is using the correct model i.e. v0.7.4-gfcd9563.

[lissyx]

Hard to help and diagnose if you dont give more context here...

[lissyx]

Quick check, the built libdeepspeech.so looks fine:

alex@portable-alex:~/tmp/deepspeech/win$ strings cpu/libdeepspeech.so|grep v0.7
v0.7.4-0-gfcd9563f
alex@portable-alex:~/tmp/deepspeech/win$ strings cuda/libdeepspeech.so|grep v0.7
v0.7.4-0-gfcd9563f
alex@portable-alex:~/tmp/deepspeech/win$ strings tflite/libdeepspeech.so|grep v0.7
v0.7.4-0-gfcd9563f
alex@portable-alex:~/tmp/deepspeech/win$ ll
total 16M
drwxrwxr-x 1 alex alex  238 juil. 18 00:15 .
drwxr-xr-x 1 alex alex 1,5K juil. 18 00:14 ..
drwxrwxr-x 1 alex alex  172 juil. 18 00:15 cpu
drwxrwxr-x 1 alex alex  172 juil. 18 00:15 cuda
-rw-rw-r-- 1 alex alex 4,3M juin  19 00:46 native_client.amd64.cpu.win.tar.xz
-rw-rw-r-- 1 alex alex  11M juin  19 00:47 native_client.amd64.cuda.win.tar.xz
-rw-rw-r-- 1 alex alex 831K juin  19 00:47 native_client.amd64.tflite.win.tar.xz
drwxrwxr-x 1 alex alex  172 juil. 18 00:15 tflite

So without more context we can't:

• know how you setup things
• know how you perform the upgrade
• know how you get this output
• know the stack you are relying on

And thus decide if there's a legit issue on our side or not ...

[lissyx]

@SanghviChirag Do you mind sharing more context ?

[reuben]

Note that that version info refers to the code, not the model. If you upgrade the model but don't upgrade the DeepSpeech package you're using, the printed version won't change.

[SanghviChirag]

Thanks, @lissyx for the response.
I've updated deepspeech version, even I uninstalled and reinstalled the latest version of deepspeech model it is showing v0.7.0.

how I setup things?
I used a virtual environment for installing package. I'm using python's 3.6.8 version (same is being used for both Mac & Linux).

pip install deepspeech

how I upgraded model?

1. First I downloaded the latest model and replaced with the old one.
2. update deepspeech version using pip install deepspeech --upgrade
3. Even I uninstalled and installed again.

How I got this output?
I followed the steps mentioned on this page: https://deepspeech.readthedocs.io/en/v0.7.4/

deepspeech --model deepspeech-0.7.4-models.pbmm --scorer deepspeech-0.7.4-models.scorer --audio audio/2830-3980-0043.wav

the stack you are relying on? I didn't get up here

[lissyx]

I've updated deepspeech version, even I uninstalled and reinstalled the latest version of deepspeech model it is showing v0.7.0.

As @reuben said, you are talking of both "deepspeech" and "deepspeech model", so we need to really be precise on what you do ...

I used a virtual environment for installing package. I'm using python's 3.6.8 version (same is being used for both Mac & Linux).
pip install deepspeech

Care to share pip install -v + pip list ?

update deepspeech version using pip install deepspeech --upgrade

This in itself should take care of running it correctly ... But without verbose upgrade log, we can't know for sure ...

Can you ensure there is no leftover libdeepspeech.so somewhere in your PATH ?

[lissyx]

python's 3.6.8 version (same is being used for both Mac & Linux).
pip install deepspeech

https://files.pythonhosted.org/packages/cf/44/958607ce1c95c5386c29e2b9c8a30f8bc797d07306d16e6ef7f7f468a464/deepspeech-0.7.4-cp36-cp36m-win_amd64.whl

So it's published, we really need more verbose logs and pip list logs, but I'm highly doubtful there's a bug on our side, it more feels like some setup issue on your side ...

[lissyx]

the stack you are relying on? I didn't get up here

Looks like you are using Python then.

[lissyx]

Please @SanghviChirag, can you share the actionable items requested ?

[SanghviChirag]

Hi Sorry, I didn't get time to work on this. I will post my results soon! Thanks

…

On Tue, Jul 21, 2020 at 8:13 PM lissyx ***@***.***> wrote: Please @SanghviChirag <https://github.com/SanghviChirag>, can you share the actionable items requested ? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3167 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACWXMCHFJOLADZ6X52ADFTDR4WSPLANCNFSM4O643JVQ> .

-- Regards, Chirag Sanghvi

[lissyx]

I didn't get time to work on this. I will post my results soon!

Please @SanghviChirag ? We can't move forward without your feedback, and it'd be important if we can assert if there is not a huge issue we missed before we issue a new stable release.

[lissyx]

@SanghviChirag We are still waiting on your infos.

[SanghviChirag]

python's 3.6.8 version (same is being used for both Mac & Linux).
pip install deepspeech

https://files.pythonhosted.org/packages/cf/44/958607ce1c95c5386c29e2b9c8a30f8bc797d07306d16e6ef7f7f468a464/deepspeech-0.7.4-cp36-cp36m-win_amd64.whl

So it's published, we really need more verbose logs and pip list logs, but I'm highly doubtful there's a bug on our side, it more feels like some setup issue on your side ...

Even after installing this, it detects v0.7.0

[SanghviChirag]

No, I downloaded pre-trained models of v0.7.4, updated deepspeech version too.

pip install -v + pip list ?
pip 18.1

pip freeze (pip list)
appdirs==1.4.4
audioread==2.1.8
cachetools==4.1.1
certifi==2020.6.20
cffi==1.14.0
chardet==3.0.4
cx-Freeze==6.1
decorator==4.4.2
deepspeech==0.7.4
google-api-core==1.22.0
google-api-python-client==1.10.0
google-auth==1.19.2
google-auth-httplib2==0.0.4
googleapis-common-protos==1.52.0
httplib2==0.18.1
idna==2.8
importlib-metadata==1.6.0
joblib==0.15.1
jsonpickle==1.4.1
librosa==0.8.0
llvmlite==0.33.0
numba==0.50.1
numpy==1.14.4
oauth2client==4.1.3
packaging==20.4
pocketsphinx==0.1.15
pooch==1.1.1
protobuf==3.12.2
py-wasapi-client==1.1.0
pyasn1==0.4.8
pyasn1-modules==0.2.8
PyAudio==0.2.11
pycparser==2.20
pydub==0.24.1
pyparsing==2.4.7
pypiwin32==223
PyQRCode==1.2.1
pyqtspinner==0.1.1
PySide2==5.14.0
pytz==2020.1
pywin32==224
requests==2.22.0
resampy==0.2.2
rsa==4.6
scikit-learn==0.23.1
scipy==1.4.1
shiboken2==5.14.0
six==1.15.0
sklearn==0.0
SoundCard==0.4.0
sounddevice==0.3.15
SoundFile==0.10.3.post1
SpeechRecognition==3.8.1
threadpoolctl==2.1.0
uritemplate==3.0.1
urllib3==1.25.9
wavio==0.0.4
win10toast==0.9
zipp==3.1.0

[lissyx]

pip install -v + pip list ?
pip 18.1

Can we be serious two seconds?
we need the installl log!

[lissyx]

Even after installing this, it detects v0.7.0

Please avoid screenshots.
Please explain how you run things, what is this shell, are you running under msys2? Under PowerShell? Under cmd.exe?
Can you ensure there is no libdeepspeech.so being hit in your PATH?
Can you strace the loading to see where the libdeepspeech.so is being loaded from?

[lissyx]

@SanghviChirag Can you please share details ?

[SanghviChirag]

I'm using Anaconda Prompt to get Linux familiarity in Windows OS.
Following are the exact steps I'm doing:

1. Create a virtualenv using python 3.6.8: python -m venv

2. Activate env: /Scripts/activate

3. pip install deepspeech (Latest version i.e. v0.8.1 is being installed.)

4. downloaded the latest pre-trained model:
   https://github.com/mozilla/DeepSpeech/releases/download/v0.8.1/deepspeech-0.8.1-models.pbmm
   https://github.com/mozilla/DeepSpeech/releases/download/v0.8.1/deepspeech-0.8.1-models.scorer

5. Script to load model:
   `from deepspeech import Model
   ds = Model('deepspeech-0.8.1-models.pbmm')

   On initializing ds:
   TensorFlow: v1.15.0-24-gceb46aae58
   DeepSpeech: v0.7.0-0-g3fbbca2b
   2020-08-13 01:29:04.289027: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
   `
   Above details are being printed on the terminal. Hope it will help.

   ds.enableExternalScorer('deepspeech-0.8.1-models.scorer')

[lissyx]

I'm using Anaconda Prompt to get Linux familiarity in Windows OS.

Well, I'm sorry but it's 1000% non supported setup. Please stick to at least pure Python virtualenv as we document.

3\. pip install deepspeech (Latest version i.e. v0.8.1 is being installed.)

Still no log of the install, so we can't verify what is happening.

5\. Script to load model:

Please verify with the deepspeech binary installed by the package, and verify it loads the correct package

5\. from deepspeech import Model

Same:

• verify it loads the proper deepspeech python module
• verify it loads the proper libdeepspeech.so

I've already asked you all those before, I'm really starting to loose patience into helping you now.

[lissyx]

It's been more than a month without news, and you are the only one reporting this weird behavior. There's something unclear, but we can't keep this issue open with no activity.