search

NervanaSystems / deepspeech

DeepSpeech neon implementation
Apache License 2.0
222 stars 69 forks source link

Error when evaluate a new trained model #30

Closed duongquangduc closed 7 years ago

duongquangduc commented 7 years ago

Hi expert, I am facing an issue when evaluate a new trained model. Could you please suggest? Here is the details: neon.util.argparser - ERROR - No read access for model file root/deepspeech/model_output.pkl None Traceback (most recent call last): File "evaluate.py", line 47, in args = parser.parse_args() File "/neon/neon/util/argparser.py", line 345, in parse_args raise IOError(err_msg) IOError: No read access for model file root/deepspeech/model_output.pkl

image

tyler-nervana commented 7 years ago

Hi, it looks like you just forgot a "/" in "/root". That's a bit of a strange error message, though, so let us know if that is not the case.

duongquangduc commented 7 years ago

Yes, it may be the cause. I changed some settings, added the "/", and now I am able to run the statement to verify the model. There are two warning messages though it can still run through. I will let me you know if there is any issue once I finish verifying the model and printing out the text corresponding to the audio file. image

In the meantime, I have a question regarding the evaluation step. In case I want to verify any audio file, which format of the audio file can be applied for this framework? Do I need to preprocess a given audio file (for example, .mp3, .wav...)?

Thank you for the support! Duc

tyler-nervana commented 7 years ago

Great to hear. The warning messages won't cause any real issues. You should be able to use any audio file that libsox can support. If you have any issues with this, you can post on the issue tracker for aeon.

duongquangduc commented 7 years ago

Hi, I am able to evaluate the sample data now. Thank you for your kind support! I will continue working on this framework. So, I hope to receive more help from you. Kind regards, Duc