{ "id": 425457, "title": "Wav2Vec2 model trained on DL Sprint Data (YellowKing)", "url": "/competitions/bengaliai-speech/discussion/425457", "author_name": "", "post_date": "2023-07-18T20:14:05.139591200Z", "votes": 25, "comment_count": 8, "views": 0, "content": "

The inference code of Wav2Vec2 model from the DL Sprint winners YellowKing is uploaded here

\n

For more details about their method please read the model paper

\n

Preprocessing and training codes

", "messages": [ { "id": "2349960", "postDate": "07/18/2023 20:14:05", "content": "

The inference code of Wav2Vec2 model from the DL Sprint winners YellowKing is uploaded here

\n

For more details about their method please read the model paper

\n

Preprocessing and training codes

", "rawMarkdown": "The inference code of Wav2Vec2 model from the [DL Sprint](https://www.kaggle.com/competitions/dlsprint/overview) winners YellowKing is uploaded [here](https://www.kaggle.com/code/reasat/yellowking-dlsprint-inference)\n\nFor more details about their method please read the [model paper](http://export.arxiv.org/abs/2209.06581)\n\n[Preprocessing and training codes](https://www.kaggle.com/competitions/bengaliai-speech/discussion/425457#2355196)", "votes": null }, { "id": "2350026", "postDate": "07/18/2023 22:46:13", "content": "

That topic should be pinned Tahsin.
\nIn general, kagglers are young and don't pay attention to \"contributors\" .

\n

All the 3 links above are very helpful. I confess that I missed many fantastic Notebooks on DL Sprint that were Not published at the beginning of the competition.

", "rawMarkdown": "That topic should be pinned Tahsin. \nIn general, kagglers are young and don't pay attention to \"contributors\" . \n\nAll the 3 links above are very helpful. I confess that I missed many fantastic Notebooks on DL Sprint that were Not published at the beginning of the competition.", "votes": null }, { "id": "2353655", "postDate": "07/21/2023 21:18:26", "content": "

This is very helpful.. thanks

", "rawMarkdown": "This is very helpful.. thanks", "votes": null }, { "id": "2354359", "postDate": "07/22/2023 11:30:05", "content": "

training code
\nhttps://www.kaggle.com/code/shahruk10/training-notebook-wav2vec2
\nhttps://huggingface.co/shahruk10/wav2vec2-xls-r-300m-bengali-commonvoice

", "rawMarkdown": "training code\nhttps://www.kaggle.com/code/shahruk10/training-notebook-wav2vec2\nhttps://huggingface.co/shahruk10/wav2vec2-xls-r-300m-bengali-commonvoice", "votes": null }, { "id": "2354853", "postDate": "07/22/2023 21:25:53", "content": "

New to this and was looking at YellowKing inference code [ here].
\nQ1: Did he/she post training notebook?

\n

I can see that he/she is loading the model?

\n

my_model_name = '../input/yellowking-dlsprint-model/YellowKing_model'

\n

Q2: Do we need to do training in our notebooks or just doing inference is enough?
\nI will be training model outside notebook.

", "rawMarkdown": "New to this and was looking at YellowKing inference code [ [here](https://www.kaggle.com/code/reasat/yellowking-dlsprint-inference)]. \nQ1: Did he/she post training notebook? \n\n I can see that he/she is loading the model? \n\n` my_model_name = '../input/yellowking-dlsprint-model/YellowKing_model'`\n\n\nQ2: Do we need to do training in our notebooks or just doing inference is enough? \nI will be training model outside notebook.", "votes": null }, { "id": "2355196", "postDate": "07/23/2023 07:03:48", "content": "

Hello, I am(was?) a member of Team YellowKing. I did make all the preprocessing, training and inference notebooks publics but didn't make a discussion post back then. You can find the links in this github repo. The reason I was loading the model from a Kaggle Dataset was because that competition has an offline inference clause.

", "rawMarkdown": "Hello, I am(was?) a member of Team YellowKing. I did make all the preprocessing, training and inference notebooks publics but didn't make a discussion post back then. You can find the links in this [github](https://github.com/Patchwork53/DLSprint2022-Champion) repo. The reason I was loading the model from a Kaggle Dataset was because that competition has an offline inference clause.", "votes": null }, { "id": "2355233", "postDate": "07/23/2023 07:42:32", "content": "

Preprocessing Notebooks

\n

Preprocessing_Stage1
\nPreprocessing_Stage2

\n

Training Notebook

\n

YellowKing_DLSprint_Training

\n

Inference Notebook

\n

YellowKing_DLSprint_Inference

\n

Technical Report

\n

Report

", "rawMarkdown": "### Preprocessing Notebooks\n[Preprocessing_Stage1](https://www.kaggle.com/code/sameen53/yellowking-dlsprint-datapreprocessingv1)\n[Preprocessing_Stage2](https://www.kaggle.com/code/sameen53/yellowking-dlsprint-datapreprocessingv2)\n\n### Training Notebook\n[YellowKing_DLSprint_Training](https://www.kaggle.com/code/sameen53/yellowking-dlsprint-training)\n\n### Inference Notebook\n[YellowKing_DLSprint_Inference](https://www.kaggle.com/code/sameen53/yellowking-dlsprint-inference)\n\n### Technical Report\n[Report](https://arxiv.org/abs/2209.06581)", "votes": null }, { "id": "2399371", "postDate": "08/20/2023 09:50:03", "content": "

just want to clarify one thing, only need to recognize the language or something else need to be done

", "rawMarkdown": "just want to clarify one thing, only need to recognize the language or something else need to be done", "votes": null }, { "id": "2399605", "postDate": "08/20/2023 13:26:34", "content": "

no, it should recognize every word. Your model will take an audio file as input and return the words as text/sentences/series of words

", "rawMarkdown": "no, it should recognize every word. Your model will take an audio file as input and return the words as text/sentences/series of words", "votes": null } ], "comments": [ { "id": 2350026, "author_name": "mpwolke", "author_url": "", "post_date": "07/18/2023 22:46:13", "content": "

That topic should be pinned Tahsin.
\nIn general, kagglers are young and don't pay attention to \"contributors\" .

\n

All the 3 links above are very helpful. I confess that I missed many fantastic Notebooks on DL Sprint that were Not published at the beginning of the competition.

", "votes": null, "replies": [] }, { "id": 2353655, "author_name": "cid007", "author_url": "", "post_date": "07/21/2023 21:18:26", "content": "

This is very helpful.. thanks

", "votes": null, "replies": [] }, { "id": 2354359, "author_name": "hengck23", "author_url": "", "post_date": "07/22/2023 11:30:05", "content": "

training code
\nhttps://www.kaggle.com/code/shahruk10/training-notebook-wav2vec2
\nhttps://huggingface.co/shahruk10/wav2vec2-xls-r-300m-bengali-commonvoice

", "votes": null, "replies": [] }, { "id": 2354853, "author_name": "robot2020", "author_url": "", "post_date": "07/22/2023 21:25:53", "content": "

New to this and was looking at YellowKing inference code [ here].
\nQ1: Did he/she post training notebook?

\n

I can see that he/she is loading the model?

\n

my_model_name = '../input/yellowking-dlsprint-model/YellowKing_model'

\n

Q2: Do we need to do training in our notebooks or just doing inference is enough?
\nI will be training model outside notebook.

", "votes": null, "replies": [ { "id": 2355196, "author_name": "sameen53", "author_url": "", "post_date": "07/23/2023 07:03:48", "content": "

Hello, I am(was?) a member of Team YellowKing. I did make all the preprocessing, training and inference notebooks publics but didn't make a discussion post back then. You can find the links in this github repo. The reason I was loading the model from a Kaggle Dataset was because that competition has an offline inference clause.

", "votes": null, "replies": [ { "id": 2355233, "author_name": "sameen53", "author_url": "", "post_date": "07/23/2023 07:42:32", "content": "

Preprocessing Notebooks

\n

Preprocessing_Stage1
\nPreprocessing_Stage2

\n

Training Notebook

\n

YellowKing_DLSprint_Training

\n

Inference Notebook

\n

YellowKing_DLSprint_Inference

\n

Technical Report

\n

Report

", "votes": null, "replies": [] } ] } ] }, { "id": 2399371, "author_name": "amishra527", "author_url": "", "post_date": "08/20/2023 09:50:03", "content": "

just want to clarify one thing, only need to recognize the language or something else need to be done

", "votes": null, "replies": [ { "id": 2399605, "author_name": "mamun18", "author_url": "", "post_date": "08/20/2023 13:26:34", "content": "

no, it should recognize every word. Your model will take an audio file as input and return the words as text/sentences/series of words

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2349960": "The inference code of Wav2Vec2 model from the [DL Sprint](https://www.kaggle.com/competitions/dlsprint/overview) winners YellowKing is uploaded [here](https://www.kaggle.com/code/reasat/yellowking-dlsprint-inference)\n\nFor more details about their method please read the [model paper](http://export.arxiv.org/abs/2209.06581)\n\n[Preprocessing and training codes](https://www.kaggle.com/competitions/bengaliai-speech/discussion/425457#2355196)", "2350026": "That topic should be pinned Tahsin. \nIn general, kagglers are young and don't pay attention to \"contributors\" . \n\nAll the 3 links above are very helpful. I confess that I missed many fantastic Notebooks on DL Sprint that were Not published at the beginning of the competition.", "2353655": "This is very helpful.. thanks", "2354359": "training code\nhttps://www.kaggle.com/code/shahruk10/training-notebook-wav2vec2\nhttps://huggingface.co/shahruk10/wav2vec2-xls-r-300m-bengali-commonvoice", "2354853": "New to this and was looking at YellowKing inference code [ [here](https://www.kaggle.com/code/reasat/yellowking-dlsprint-inference)]. \nQ1: Did he/she post training notebook? \n\n I can see that he/she is loading the model? \n\n` my_model_name = '../input/yellowking-dlsprint-model/YellowKing_model'`\n\n\nQ2: Do we need to do training in our notebooks or just doing inference is enough? \nI will be training model outside notebook.", "2355196": "Hello, I am(was?) a member of Team YellowKing. I did make all the preprocessing, training and inference notebooks publics but didn't make a discussion post back then. You can find the links in this [github](https://github.com/Patchwork53/DLSprint2022-Champion) repo. The reason I was loading the model from a Kaggle Dataset was because that competition has an offline inference clause.", "2355233": "### Preprocessing Notebooks\n[Preprocessing_Stage1](https://www.kaggle.com/code/sameen53/yellowking-dlsprint-datapreprocessingv1)\n[Preprocessing_Stage2](https://www.kaggle.com/code/sameen53/yellowking-dlsprint-datapreprocessingv2)\n\n### Training Notebook\n[YellowKing_DLSprint_Training](https://www.kaggle.com/code/sameen53/yellowking-dlsprint-training)\n\n### Inference Notebook\n[YellowKing_DLSprint_Inference](https://www.kaggle.com/code/sameen53/yellowking-dlsprint-inference)\n\n### Technical Report\n[Report](https://arxiv.org/abs/2209.06581)", "2399371": "just want to clarify one thing, only need to recognize the language or something else need to be done", "2399605": "no, it should recognize every word. Your model will take an audio file as input and return the words as text/sentences/series of words" }, "source": "meta" }