Guideline for Specific Version of Deepspeech

(Hafsa Farooq) #1

Hi everyone!

There is constantly changes in Deepspeech versions. I installed Deepspeech 0.2.1 alpha in September 2018. Up till now, Deepspeech 0.4.1. has been released.
I am wondering how to keep track of a specific version.
Its becoming a big confusion. Kindly help.
Or Anything better which can be recommended to developers.
Thank you!

(kdavis) #2

You can always find which release is the most recent by looking at the GitHub DeepSpeech releases page.

Alternatively you can just update, say, your python package

pip3 install deepspeech --upgrade

to get the latest version.

(Hafsa Farooq) #3

What to do to find the help of a specific version ?

(Lissyx) #4

What help are you looking for ?

(Hafsa Farooq) #5

I am using DeepSpeech 0.2.1 alpha 0. For English Language, it doesn’t decode English sentences with any spaces.
For Urdu Language, it stuck on the very first character of my vocabulary.
I want to see the Github DeepSpeech release page for DeepSpeech master 0.2.1 . But the latest version come out with ctc decoder, which I had not been using for previous versions.

(Lissyx) #6

We never published a stable 0.2.1, we moved to 0.3.0, because of a change breaking backward compatibility.

(Lissyx) #7
(Hafsa Farooq) #8

Thank you! What could be problem that pre-trained model doesn’t decode sentences with spaces.

(Hafsa Farooq) #9

Mostly it decodes correctly but without any spaces.

(Lissyx) #10

This is a bug we fixed by moving to ctcdecoder

(Hafsa Farooq) #11

This is what I was figuring it out. Thank you for the help. :smile:

(Hafsa Farooq) #12

Can somebody share the complete training guide of DeepSpeech 0.4.1 ?
guideline with data files format and how many files are needed ?
Like, Does it need vocab.txt, binary trie file ?
Or everything to train from DeepSpeech 0.4.1
thank you!

(Hafsa Farooq) #13

Traceback (most recent call last):
File “./DeepSpeech.py”, line 11, in
import evaluate
File “/home/rc/Desktop/0.4.1/DeepSpeech-master/evaluate.py”, line 17, in
from ds_ctcdecoder import ctc_beam_search_decoder_batch, Scorer
ImportError: No module named ds_ctcdecode
I am having this error though I install ctc decoder. using
pip3 install $(python3 util/taskcluster.py --decoder)
I am also missing trie file in DeepSpeech 0.4.1 I don’t know how to generate it.
Thank you for the help.

(Lissyx) #14

Well, the error shows you don’t have the module installed … Maybe messing around your virtualenv ?

Please refer to the documentation.

(Hafsa Farooq) #15

Hi, Does any version of Deepspeech works on Ubuntu 14.04 and supports cuda 8 rather than cuda 9?
I searched but I couldn’t get the specific answer.