Where I can get pre-trained model for version 5

Deep Speech
#1

Hello! I wanna use deepSpeech with
-e Extended output, shows word timings as CSV (word, start time, duration)
So I download native_client from https://github.com/mozilla/DeepSpeech/releases/tag/v0.5.0-alpha.5

And latest pre-trained model deepspeech-0.4.1-models.tar.gz
But when I try to run code I got this error

TensorFlow: v1.13.1-10-g3e0cc5374d
DeepSpeech: v0.5.0-alpha.5-0-g15adf00
2019-04-10 13:07:19.753579: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2019-04-10 13:07:19.763497: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel ('op: "UnwrapDatasetVariant" device_type: "GPU" host_memory_arg: "input_handle" host_memory_arg: "output_handle"') for unknown op: UnwrapDatasetVariant
2019-04-10 13:07:19.763526: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel ('op: "UnwrapDatasetVariant" device_type: "CPU"') for unknown op: UnwrapDatasetVariant
2019-04-10 13:07:19.763553: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel ('op: "WrapDatasetVariant" device_type: "GPU" host_memory_arg: "input_handle" host_memory_arg: "output_handle"') for unknown op: WrapDatasetVariant
2019-04-10 13:07:19.763574: E tensorflow/core/framework/op_kernel.cc:1325] OpKernel ('op: "WrapDatasetVariant" device_type: "CPU"') for unknown op: WrapDatasetVariant
Specified model file version (0) is incompatible with minimum version supported by this client (1). See https://github.com/mozilla/DeepSpeech/#model-compatibility for more information
Could not create model.

Can you plz help me ? Thanks.

#2

Currently there’s no pre-trained model available for the 0.5 alpha builds.

#3

Can you plz give me some roughly estimate when it will be done?

#4

No, we don’t have any ETA.

#5

You can export a model from 0.4.1 checkpoint.

1 Like
#6

eggonlea when you suggest exporting a model from 0.4.1 is this because you know that 0.4.1 exports are compatible with v0.5.0-alpha.x runtime?

I would’ve thought the files in, say deepspeech-0.4.1-models.tar.gz are already in exported, not checkpoint, form…

Or… ah! I think you mean that you can load a model from 0.4.1 checkpoint into a v0.5.0-alpha.x DeepSpeech.py and then export it?

Just wondering, forgive me, but have you actually tried that?

#7

@eggonlea and anyone else thanks for your comments. FWIW I have now tried to load from a deepspeech_gpu-0.4.0a0 checkpoint into v.5.0-alpha.5 version of DeepSpeech.py – so I could then export out output_graph.pbmm.

However I can report, that this fails with the a Key layer_1/bias not found in checkpoint error.

I’ll note that you did specify a 0.4.1 checkpoint not a 0.4.0 so perhaps that’s why it didn’t work.

Anyway for what it’s worth this doesn’t work, fails as follows for me.

Guess we’re going to have to re-train our model on v0.5-alpha :frowning:

python3 DeepSpeech/DeepSpeech.py  \
  --alphabet_config_path 'model/full_model/alphabet.txt' \
  --lm_binary_path 'model/full_model/lm.pointers.binary' \
  --lm_trie_path 'model/full_model/pointers.trie' \
  --checkpoint_dir "model/full_model/checkpoint" \
  --log_level 0 \
  --notrain \
  --notest \
  --export_dir "model/full_model/export"

<snip>

I Exporting the model...
WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.
Instructions for updating:
Colocations handled automatically by placer.
<snip>
2019-05-03 07:50:30.579242: W tensorflow/core/framework/op_kernel.cc:1401] OP_REQUIRES failed at save_restore_v2_ops.cc:184 : Not found: Key layer_1/bias not found in checkpoint
Traceback (most recent call last):
  File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py", line 1334, in _do_call
    return fn(*args)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py", line 1319, in _run_fn
    options, feed_dict, fetch_list, target_list, run_metadata)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py", line 1407, in _call_tf_sessionrun
    run_metadata)
tensorflow.python.framework.errors_impl.NotFoundError: Key layer_1/bias not found in checkpoint
         [[{{node save/RestoreV2}}]]
#8

There’s another parameter that @reuben reminded us about: with the pipeline changes that brings the key not found error also comes features computation changes, i.e., the way audio is being used by the network changed. So you can’t just re-use the checkpoint.

As pointed in several topics / issues, we will provide a newly-compatible trained model / checkpoints for this version as soon as we can.

#9

Yes. Thanks, heaps @lissyx… we actually have our own models for a different language so no worries there.

Guess we were just trying to be lazy and get access to your great new word timing changes without having to re-train our model (There was a rumor going around the office that this might be possible since the core network hadn’t changed ;). No worries though we’ll re-train :wink: