I am new to DeepSpeech and Ubuntu. I am using DeepSpeech version 0.7.4 and was able to fine tune the model with my own data but got stuck while converting generated graph file to memory mapped format. I look at few bugs already created in this regard but couldn’t understand a lot because of maybe my experience with ubuntu. I am working on google colab and I ran following command(as mentioned in here) and I get following error.
Please help me out here.
Thanks a lot in advance.
lissyx
((slow to reply) [NOT PROVIDING SUPPORT])
2
which error? you don’t give any.
lissyx
((slow to reply) [NOT PROVIDING SUPPORT])
3
All you have to do is get into the DeepSpeech directory (assuming you cloned it from github) and do the following:
cd DeepSpeech # get into the directory you cloned
python3 util/taskcluster.py --source tensorflow --artifact convert_graphdef_memmapped_format --branch r1.15 --target .
chmod +x convert_graphdef_memmapped_format # make it executable
./convert_graphdef_memmapped_format --in_graph=path/to/pb/model/you/created.pb --out_graph=/destination/directory/output_graph.pbmm
Your --in_graph is the file that was generated after you finished training a model (as you said you did via fine-tuning)
You can name the --out_graph file whatever you want as long as you specify it with a .pbmm file extension.