Does not work

[KindSpidey]

Hey, everyone! Can you tell me, please, how to make this package work?

I follow the intruction:

• do pip install audiosr==0.0.3
• installation is success

Then trying to apply library to my audiofile:

D:\SONGer> audiosr -i track.mp3

and got error C:\Users\my_name\AppData\Local\Programs\Python\Python311\python.exe: No module named audiosr.__main__; 'audiosr' is a package and cannot be directly executed

Well, then I tried to read py files of project and use it like library in code by doing this:

import audiosr
from audiosr.latent_diffusion.models.ddpm import DDPM, LatentDiffusion

input_file = "track.mp3"

audiosr.super_resolution(input_file=input_file, latent_diffusion=LatentDiffusion)

but this got me to error TypeError: LatentDiffusion.generate_batch() missing 1 required positional argument: 'batch' if do this -audiosr.super_resolution(input_file=input_file, latent_diffusion=LatentDiffusion(DDPM)) than this

assert self.num_timesteps_cond <= kwargs["timesteps"]
                                      ~~~~~~^^^^^^^^^^^^^
KeyError: 'timesteps'

How to make it work locally?

[falseywinchnet]

send boobs

import os import torch import logging from audiosr import super_resolution, build_model, save_wave, get_time, read_list import argparse

os.environ["TOKENIZERS_PARALLELISM"] = "true" matplotlib_logger = logging.getLogger('matplotlib') matplotlib_logger.setLevel(logging.WARNING) torch.set_float32_matmul_precision("high") audiosr = build_model(model_name="speech", device="auto")

waveform = super_resolution( audiosr, "C:\pathtofile\./filename.wav", #use a sampling rate less than 42! recommend 6k mono. use a file less than 10s long!. seed=42, guidance_scale=3.5, ddim_steps=50, latent_t_per_second=12.8 )

save_wave(waveform, "C:\pathtofile\./", name="test", samplerate=48000) #note: do not put .wav at end of file name, it will do this automatically and grumble if you do it

[ThomasFan1945]

send boobs

import os import torch import logging from audiosr import super_resolution, build_model, save_wave, get_time, read_list import argparse

os.environ["TOKENIZERS_PARALLELISM"] = "true" matplotlib_logger = logging.getLogger('matplotlib') matplotlib_logger.setLevel(logging.WARNING) torch.set_float32_matmul_precision("high") audiosr = build_model(model_name="speech", device="auto")

waveform = super_resolution( audiosr, "C:\pathtofile./filename.wav", #use a sampling rate less than 42! recommend 6k mono. use a file less than 10s long!. seed=42, guidance_scale=3.5, ddim_steps=50, latent_t_per_second=12.8 )

save_wave(waveform, "C:\pathtofile./", name="test", samplerate=48000) #note: do not put .wav at end of file name, it will do this automatically and grumble if you do it

Hey, my dude. Can you please show us in video form? I have a better understanding by watching videos, thanks.

[falseywinchnet]

you can just use your nominal sampling rate and it will work fine as long as you highpass your data

On Thu, Sep 21, 2023 at 4:12 PM ThomasFan1945 @.***> wrote:

send boobs

import os import torch import logging from audiosr import super_resolution, build_model, save_wave, get_time, read_list import argparse

os.environ["TOKENIZERS_PARALLELISM"] = "true" matplotlib_logger = logging.getLogger('matplotlib') matplotlib_logger.setLevel(logging.WARNING) torch.set_float32_matmul_precision("high") audiosr = build_model(model_name="speech", device="auto")

waveform = super_resolution( audiosr, "C:\pathtofile./filename.wav", #use a sampling rate less than 42! recommend 6k mono. use a file less than 10s long!. seed=42, guidance_scale=3.5, ddim_steps=50, latent_t_per_second=12.8 )

save_wave(waveform, "C:\pathtofile./", name="test", samplerate=48000)

note: do not put .wav at end of file name, it will do this automatically

and grumble if you do it

Hey, my dude. Can you please show us in video form? I have a better understanding by watching videos, thanks.

— Reply to this email directly, view it on GitHub https://github.com/haoheliu/versatile_audio_super_resolution/issues/17#issuecomment-1730308893, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUYJQQ5CWJK5OASVGIJ3P6DX3SUS5ANCNFSM6AAAAAA5BED62M . You are receiving this because you commented.Message ID: @.*** com>

[ThomasFan1945]

you can just use your nominal sampling rate and it will work fine as long as you highpass your data … On Thu, Sep 21, 2023 at 4:12 PM ThomasFan1945 @.> wrote: send boobs import os import torch import logging from audiosr import super_resolution, build_model, save_wave, get_time, read_list import argparse os.environ["TOKENIZERS_PARALLELISM"] = "true" matplotlib_logger = logging.getLogger('matplotlib') matplotlib_logger.setLevel(logging.WARNING) torch.set_float32_matmul_precision("high") audiosr = build_model(model_name="speech", device="auto") waveform = super_resolution( audiosr, "C:\pathtofile./filename.wav", #use a sampling rate less than 42! recommend 6k mono. use a file less than 10s long!. seed=42, guidance_scale=3.5, ddim_steps=50, latent_t_per_second=12.8 ) save_wave(waveform, "C:\pathtofile./", name="test", samplerate=48000) #note: do not put .wav at end of file name, it will do this automatically and grumble if you do it Hey, my dude. Can you please show us in video form? I have a better understanding by watching videos, thanks. — Reply to this email directly, view it on GitHub <#17 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUYJQQ5CWJK5OASVGIJ3P6DX3SUS5ANCNFSM6AAAAAA5BED62M . You are receiving this because you commented.Message ID: @. com>

This is confusing. Please do a video tutorial.

[KindSpidey]

send boobs

import os import torch import logging from audiosr import super_resolution, build_model, save_wave, get_time, read_list import argparse

os.environ["TOKENIZERS_PARALLELISM"] = "true" matplotlib_logger = logging.getLogger('matplotlib') matplotlib_logger.setLevel(logging.WARNING) torch.set_float32_matmul_precision("high") audiosr = build_model(model_name="speech", device="auto")

waveform = super_resolution( audiosr, "C:\pathtofile./filename.wav", #use a sampling rate less than 42! recommend 6k mono. use a file less than 10s long!. seed=42, guidance_scale=3.5, ddim_steps=50, latent_t_per_second=12.8 )

save_wave(waveform, "C:\pathtofile./", name="test", samplerate=48000) #note: do not put .wav at end of file name, it will do this automatically and grumble if you do it

your code starts download 6GB model speech. I guees, it is not cool

[falseywinchnet]

you can just use your nominal sampling rate and it will work fine as long as you highpass your data … On Thu, Sep 21, 2023 at 4:12 PM ThomasFan1945 @._> wrote: send boobs import os import torch import logging from audiosr import super_resolution, build_model, save_wave, get_time, read_list import argparse os.environ["TOKENIZERS_PARALLELISM"] = "true" matplotlib_logger = logging.getLogger('matplotlib') matplotlib_logger.setLevel(logging.WARNING) torch.set_float32_matmul_precision("high") audiosr = build_model(model_name="speech", device="auto") waveform = super_resolution( audiosr, "C:\pathtofile./filename.wav", #use a sampling rate less than 42! recommend 6k mono. use a file less than 10s long!. seed=42, guidance_scale=3.5, ddim_steps=50, latent_t_per_second=12.8 ) save_wave(waveform, "C:\pathtofile./", name="test", samplerate=48000) #note: do not put .wav at end of file name, it will do this automatically and grumble if you do it Hey, my dude. Can you please show us in video form? I have a better understanding by watching videos, thanks. — Reply to this email directly, view it on GitHub <#17 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUYJQQ5CWJK5OASVGIJ3P6DX3SUS5ANCNFSM6AAAAAA5BED62M . You are receiving this because you commented.Message ID: _@_._ com>

This is confusing. Please do a video tutorial.

pay me?

[madwurmz]

@falseywinchnet I will literately pay somebody to get this working! 🗣️ It feels so promising and highly valuable , but I can't get this to work, I tried a lot of things many different errors!

[falseywinchnet]

i would be glad to take your money here is my recommendation: use miniforge python, not other release(not anaconda, etc)

pip install audiosr

in python IDE or ipy notebook: import os import torch import logging from audiosr import super_resolution, build_model, save_wave, get_time, read_list import argparse

os.environ["TOKENIZERS_PARALLELISM"] = "true" matplotlib_logger = logging.getLogger('matplotlib') matplotlib_logger.setLevel(logging.WARNING) torch.set_float32_matmul_precision("high")

audiosr = build_model(model_name="speech", device="auto") #this task will download a 6gb file. be sure you have the hard drive space, and enough RAM.

if you are using cuda, you will need at least 8gb VRAM!

waveform = super_resolution( audiosr, "C:\pathtofile./filename.wav", seed=42, guidance_scale=3.5, ddim_steps=50, latent_t_per_second=12.8 ) #note: just use your ordinary audio and low-pass filter it before applying the network. just like inpainting a picture!

save_wave(waveform, "C:\pathtofile./", name="test", samplerate=48000) #note: do not put .wav at end of file name, audiosr does this automatically, does not like it if you do this

[madwurmz]

@falseywinchnet Thanks! That sounds helpful! Do you mean I need to do things like import os necessarily because the bare repo is handling that insufficiently?

We're in the wrong channel to chat but initially I had it kind of installed with anaconda. I had to search where it was installed , not in my preferred directory, this repo installs files then hides them deep in the windows 10 folders, I found it in some Scripts folder.. a few attempts gave errors related to permissions and filenames, then I tried to convert a batch, it downloaded the 6gb model and it filled 10gb vram and then it froze. I have a lot of samples to convert , curious to the quality . I also could imagine this tool working inside bark and other ai generators 🎵

[falseywinchnet]

your not paying me enough

On Sat, Sep 23, 2023 at 10:36 PM madwurmz @.***> wrote:

@falseywinchnet https://github.com/falseywinchnet Thanks! That sounds helpful! Do you mean I need to do things like import os necessarily because the bare repo is handling that insufficiently?

We're in the wrong channel to chat but initially I had it kind of installed with anaconda. I had to search where it was installed , not in my preferred directory, this repo installs files then hides them deep in the windows 10 folders, I found it in some Scripts folder.. a few attempts gave errors related to permissions and filenames, then I tried to convert a batch, it downloaded the 6gb model and it filled 10gb vram and then it froze. I have a lot of samples to convert , curious to the quality . I also could imagine this tool working inside bark and other ai generators 🎵

— Reply to this email directly, view it on GitHub https://github.com/haoheliu/versatile_audio_super_resolution/issues/17#issuecomment-1732467502, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUYJQQ275JWCYJI6X2GEPBLX36TEXANCNFSM6AAAAAA5BED62M . You are receiving this because you were mentioned.Message ID: @.*** com>

[madwurmz]

@falseywinchnet ohlala 🛩️ I have 1000's of samples, if this tool is working, it is really worth something! I can give anybody who can make this working, a ko-fi, or I can pay you with a compliment, pay respect or pay with a personalized rap song 💯

[falseywinchnet]

all you have to do is learn a little bit of python

On Sun, Sep 24, 2023 at 8:53 AM madwurmz @.***> wrote:

@falseywinchnet https://github.com/falseywinchnet ohlala 🛩️ I have 1000's of samples, if this tool is working, it is really worth something! I can give anybody who can this working, a ko-fi, or I can pay you with a compliment, pay respect or pay with a personalized rap song 💯

— Reply to this email directly, view it on GitHub https://github.com/haoheliu/versatile_audio_super_resolution/issues/17#issuecomment-1732575362, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUYJQQ53Y6UKWJA4QTTTDXLX4A3NXANCNFSM6AAAAAA5BED62M . You are receiving this because you were mentioned.Message ID: @.*** com>

[madwurmz]

@falseywinchnet the command you given is wrong , no way a dot is needed in the directory?? and it says you mix windows with unix style formating when you suggest this: C:\pathtofile./filename.wav

can you suggest how to rewrite it to just ran any batch?

Im at the part it says import os, so I'm very serious! but I have anaconda and python 3.10 and that always working fine, are you sponsored by miniforge?

I will pay real money to get this is working! dont worry about that! 🥇 or I write you a dis rap song if you keep on being so sarcastic! 🍡

edit: reading other reports saying this needs 16 gb vram ?! thats why it wont even work for me sitting on 11gb... 😞

i would be glad to take your money here is my recommendation: use miniforge python, not other release(not anaconda, etc)

pip install audiosr

in python IDE or ipy notebook: import os import torch import logging from audiosr import super_resolution, build_model, save_wave, get_time, read_list import argparse

os.environ["TOKENIZERS_PARALLELISM"] = "true" matplotlib_logger = logging.getLogger('matplotlib') matplotlib_logger.setLevel(logging.WARNING) torch.set_float32_matmul_precision("high")

audiosr = build_model(model_name="speech", device="auto") #this task will download a 6gb file. be sure you have the hard drive space, and enough RAM. #if you are using cuda, you will need at least 8gb VRAM!

waveform = super_resolution( audiosr, "C:\pathtofile./filename.wav", seed=42, guidance_scale=3.5, ddim_steps=50, latent_t_per_second=12.8 ) #note: just use your ordinary audio and low-pass filter it before applying the network. just like inpainting a picture!

save_wave(waveform, "C:\pathtofile./", name="test", samplerate=48000) #note: do not put .wav at end of file name, audiosr does this automatically, does not like it if you do this