We are limited to 2 submissions to select for final scoring.
\n\nDoes this mean 2 submissions for the regular competition and 2 different submissions for the Tensorflow special prize?
\n\nIt seems otherwise we will have to choose whether to compete in the regular or the special competition, or be at a disadvantage by submitting only 1 submission for each.
\n\nCompetition organizers, could you clarify the process for submitting for the special prize?
\n\nI would like to be able to compete in both areas
", "messages": [ { "id": "266768", "postDate": "01/09/2018 17:08:19", "content": "We are limited to 2 submissions to select for final scoring.
\n\nDoes this mean 2 submissions for the regular competition and 2 different submissions for the Tensorflow special prize?
\n\nIt seems otherwise we will have to choose whether to compete in the regular or the special competition, or be at a disadvantage by submitting only 1 submission for each.
\n\nCompetition organizers, could you clarify the process for submitting for the special prize?
\n\nI would like to be able to compete in both areas
", "rawMarkdown": "We are limited to 2 submissions to select for final scoring.\n\nDoes this mean 2 submissions for the regular competition and 2 different submissions for the Tensorflow special prize? \n\nIt seems otherwise we will have to choose whether to compete in the regular or the special competition, or be at a disadvantage by submitting only 1 submission for each.\n\nCompetition organizers, could you clarify the process for submitting for the special prize?\n\nI would like to be able to compete in both areas", "votes": null }, { "id": "266784", "postDate": "01/09/2018 18:23:35", "content": "Also have the same doubt here.
", "rawMarkdown": "Also have the same doubt here.", "votes": null }, { "id": "266786", "postDate": "01/09/2018 18:25:47", "content": "I believe I have read in another post that submission for the special prize will be done separately and that will be ready for the end of the challenge.
", "rawMarkdown": "I believe I have read in another post that submission for the special prize will be done separately and that will be ready for the end of the challenge.", "votes": null }, { "id": "266798", "postDate": "01/09/2018 19:45:38", "content": "same.
", "rawMarkdown": "same.", "votes": null }, { "id": "266836", "postDate": "01/09/2018 23:31:55", "content": "Another question regarding the special prize. Is there an example anywhere of how to make the graph like they want it? I don't get how you would define an argument with the name \"decoded_sample_data:0\"? Also they say the outputs should just be the softmax, but how will they know which neuron corresponds to which word (for example that the first gives the probability for 'on')? AND In their tutorial they have this script for creating a graph (different arguments) where they use from tensorflow.contrib.framework.python.ops import audio_ops as contrib_audio, but this doesn't exist in tensorflow 1.4 (so it seems you have to write your own spectogram function??)!
The script you linked contains a good example of how to create the binary graph. First, the 'decoded_sample_data:0' is simply a tf placeholder with the name 'decoded_sample_data', as you can see in line 85 of the script (0 is simply the index of the op). For the second question, there this other script (and also, its C equivalent) where the graph is actually used to predict. The problem of the ordering of the softmax outputs is resolved by passing a text file, containing in order the labels (one per line). For the audio_ops question I'm using tensorflow 1.4 and everything is running fine
", "rawMarkdown": "The script you linked contains a good example of how to create the binary graph. First, the 'decoded_sample_data:0' is simply a tf placeholder with the name 'decoded_sample_data', as you can see in line 85 of the script (0 is simply the index of the op). For the second question, there this other [script][1] (and also, its C equivalent) where the graph is actually used to predict. The problem of the ordering of the softmax outputs is resolved by passing a text file, containing in order the labels (one per line). For the audio_ops question I'm using tensorflow 1.4 and everything is running fine \n\n\n [1]: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/speech_commands/label_wav.py", "votes": null }, { "id": "266890", "postDate": "01/10/2018 04:23:03", "content": "I used code from here and it just worked.
", "rawMarkdown": "I used code from [here][1] and it just worked.\n\n\n [1]: https://github.com/ARM-software/ML-KWS-for-MCU", "votes": null }, { "id": "267236", "postDate": "01/10/2018 22:17:22", "content": "Answered here:
\n\nhttps://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/47246
", "rawMarkdown": "Answered here:\n\nhttps://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/47246", "votes": null } ], "comments": [ { "id": 266784, "author_name": "xiaozhouwang", "author_url": "", "post_date": "01/09/2018 18:23:35", "content": "Also have the same doubt here.
", "votes": null, "replies": [] }, { "id": 266786, "author_name": "ironbar", "author_url": "", "post_date": "01/09/2018 18:25:47", "content": "I believe I have read in another post that submission for the special prize will be done separately and that will be ready for the end of the challenge.
", "votes": null, "replies": [] }, { "id": 266798, "author_name": "left13", "author_url": "", "post_date": "01/09/2018 19:45:38", "content": "same.
", "votes": null, "replies": [] }, { "id": 266836, "author_name": "nimitz14", "author_url": "", "post_date": "01/09/2018 23:31:55", "content": "Another question regarding the special prize. Is there an example anywhere of how to make the graph like they want it? I don't get how you would define an argument with the name \"decoded_sample_data:0\"? Also they say the outputs should just be the softmax, but how will they know which neuron corresponds to which word (for example that the first gives the probability for 'on')? AND In their tutorial they have this script for creating a graph (different arguments) where they use from tensorflow.contrib.framework.python.ops import audio_ops as contrib_audio, but this doesn't exist in tensorflow 1.4 (so it seems you have to write your own spectogram function??)!
The script you linked contains a good example of how to create the binary graph. First, the 'decoded_sample_data:0' is simply a tf placeholder with the name 'decoded_sample_data', as you can see in line 85 of the script (0 is simply the index of the op). For the second question, there this other script (and also, its C equivalent) where the graph is actually used to predict. The problem of the ordering of the softmax outputs is resolved by passing a text file, containing in order the labels (one per line). For the audio_ops question I'm using tensorflow 1.4 and everything is running fine
", "votes": null, "replies": [] }, { "id": 266890, "author_name": "bsp2020", "author_url": "", "post_date": "01/10/2018 04:23:03", "content": "I used code from here and it just worked.
", "votes": null, "replies": [] } ] }, { "id": 267236, "author_name": "inversion", "author_url": "", "post_date": "01/10/2018 22:17:22", "content": "Answered here:
\n\nhttps://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/47246
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "266768": "We are limited to 2 submissions to select for final scoring.\n\nDoes this mean 2 submissions for the regular competition and 2 different submissions for the Tensorflow special prize? \n\nIt seems otherwise we will have to choose whether to compete in the regular or the special competition, or be at a disadvantage by submitting only 1 submission for each.\n\nCompetition organizers, could you clarify the process for submitting for the special prize?\n\nI would like to be able to compete in both areas", "266784": "Also have the same doubt here.", "266786": "I believe I have read in another post that submission for the special prize will be done separately and that will be ready for the end of the challenge.", "266798": "same.", "266836": "Another question regarding the special prize. Is there an example anywhere of how to make the graph like they want it? I don't get how you would define an argument with the name \"decoded_sample_data:0\"? Also they say the outputs should just be the softmax, but how will they know which neuron corresponds to which word (for example that the first gives the probability for 'on')? AND In their tutorial they have [this](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/speech_commands/freeze.py) script for creating a graph (different arguments) where they use `from tensorflow.contrib.framework.python.ops import audio_ops as contrib_audio`, but this doesn't exist in tensorflow 1.4 (so it seems you have to write your own spectogram function??)!", "266865": "The script you linked contains a good example of how to create the binary graph. First, the 'decoded_sample_data:0' is simply a tf placeholder with the name 'decoded_sample_data', as you can see in line 85 of the script (0 is simply the index of the op). For the second question, there this other [script][1] (and also, its C equivalent) where the graph is actually used to predict. The problem of the ordering of the softmax outputs is resolved by passing a text file, containing in order the labels (one per line). For the audio_ops question I'm using tensorflow 1.4 and everything is running fine \n\n\n [1]: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/speech_commands/label_wav.py", "266890": "I used code from [here][1] and it just worked.\n\n\n [1]: https://github.com/ARM-software/ML-KWS-for-MCU", "267236": "Answered here:\n\nhttps://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/47246" }, "source": "meta" }