search

bill9800 / speech_separation

Include some core functions and model to handle speech separation
MIT License
153 stars 61 forks source link

not working on colab #28

Open faheem-khaskheli opened 3 years ago

faheem-khaskheli commented 3 years ago

cannot download youtube videos or audio

mwaseemrandhawa commented 3 years ago

Same issue. are you resolved it?

faheem-khaskheli commented 3 years ago

yes, i solved the issues, have to install some modules and change path of files when using these in single jupyter notebook.

mwaseemrandhawa commented 3 years ago

can you please share a code with me?

faheem-khaskheli commented 3 years ago

i will upload it on github today.

mwaseemrandhawa commented 3 years ago

thank you so much I really appreciate it. Have you successfully executed this code I mean model on the dataset?

On Wed, Oct 7, 2020 at 6:57 AM faheem khaskheli notifications@github.com wrote:

i will upload it on github today.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/bill9800/speech_separation/issues/28#issuecomment-704954311, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHHBFENWBUFNCKBTLRQP2V3SJRXVBANCNFSM4SF3OUUQ .

-- Waseem Randhawa, Department of Computer Science, NCBA&E Cell +92-3038232954

faheem-khaskheli commented 3 years ago

yes, but downloading dataset on colab is taking hours

mwaseemrandhawa commented 3 years ago

If you have a dataset and running model on colab please share your folder with me? mwaseemrahhawa@gmail.com

On Wed, Oct 7, 2020 at 7:05 AM faheem khaskheli notifications@github.com wrote:

yes, but downloading dataset on colab is taking hours

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/bill9800/speech_separation/issues/28#issuecomment-704959687, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHHBFEILFHFCZWSOOB54DUDSJRYUFANCNFSM4SF3OUUQ .

-- Waseem Randhawa, Department of Computer Science, NCBA&E Cell +92-3038232954

faheem-khaskheli commented 3 years ago

i download 100 audio and videos to test the model

mwaseemrandhawa commented 3 years ago

This is enough. I also needed to test this model using pretrained model

On Wed, Oct 7, 2020 at 7:12 AM faheem khaskheli notifications@github.com wrote:

i download 100 audio and videos to test the model

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/bill9800/speech_separation/issues/28#issuecomment-704964017, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHHBFELSDUQZCWYXVRXO3MTSJRZM3ANCNFSM4SF3OUUQ .

-- Waseem Randhawa, Department of Computer Science, NCBA&E Cell +92-3038232954

faheem-khaskheli commented 3 years ago

this model is not pretrained, we have to train it using whole dataset. do you have pretrained weights for this model?

mwaseemrandhawa commented 3 years ago

check this link it's already a saved model. https://github.com/bill9800/speech_separation/tree/master/model/model_v1/saved_models_AO

On Wed, Oct 7, 2020 at 7:16 AM faheem khaskheli notifications@github.com wrote:

this model is not pretrained, we have to train it using whole dataset. do you have pretrained weights for this model?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/bill9800/speech_separation/issues/28#issuecomment-704966751, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHHBFEPMLR6E734NIVQS5SDSJRZ4JANCNFSM4SF3OUUQ .

-- Waseem Randhawa, Department of Computer Science, NCBA&E Cell +92-3038232954

faheem-khaskheli commented 3 years ago

according to readme, model v1 is audio only. model v2 is both video and audio. these weights belong to v1 audio only model

mwaseemrandhawa commented 3 years ago

Can we talk on whats app? 03038232954

On Wed, Oct 7, 2020 at 7:24 AM faheem khaskheli notifications@github.com wrote:

according to readme, model v1 is audio only. model v2 is both video and audio. these weights belong to v1 audio only model

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/bill9800/speech_separation/issues/28#issuecomment-704972072, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHHBFEPQTD4OQG646LDFN6DSJR22FANCNFSM4SF3OUUQ .

-- Waseem Randhawa, Department of Computer Science, NCBA&E Cell +92-3038232954

faheemkhaskheli9 commented 3 years ago

here, i uploaded file that i used on colab. https://github.com/faheemkhaskheli9/speech_separation if you have any issue then contact me here, i can fix it.

mwaseemrandhawa commented 3 years ago

thank you so much. I really appreciate it.

On Wed, Oct 7, 2020 at 9:01 AM Faheem Khaskheli notifications@github.com wrote:

here, i uploaded file that i used on colab. https://github.com/faheemkhaskheli9/speech_separation if you have any issue then contact me here, i can fix it.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/bill9800/speech_separation/issues/28#issuecomment-705034983, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHHBFEIXYWQYEP2WEKQAMMLSJSGHHANCNFSM4SF3OUUQ .

-- Waseem Randhawa, Department of Computer Science, NCBA&E Cell +92-3038232954

koravadi commented 3 years ago

Since we are using the bidirectional LSTM in the network, can this be used for the realtime application? or this works only for the recorded data?

faheem-khaskheli commented 3 years ago

it can be used on real time data, you just have to split the input stream into chunks, like get 3 second data and feed it to model, then another 3 second data. you have to wait for 3 second so that you can collect 3 second recording. that will be lag in system.

koravadi commented 3 years ago

@faheem-khaskheli thanks for clarification!

koravadi commented 3 years ago

@faheem-khaskheli I have one more question. Why it needs to be 3 sec why can't it be 1 sec? is there any parameter defines this 3 sec interval in the network?

faheemkhaskheli9 commented 3 years ago

no reason, you can make it 1 second but you need some useful data to do task so long will be better then 1 sec or less then 1 sec

SutirthaChakraborty commented 2 years ago

Thanks @faheemkhaskheli9 I tried to replicate your colab code. The last part 'evaluate the model and generate the prediction' is not working. I am getting silent audio as output. Any help ?