Reduce noise in stereo

[olawalejuwonm]

Hi.

Thanks for the awesome project.

The output of the sound is in mono (one channel), can there be option to set this or default it to multiple channel?

[nuniz]

The input to noisereduce is a 2-dimensional matrix, so you can input stereo into the algorithm and obtain stereo output. However, the noise estimation is not shared between the channels.

[olawalejuwonm]

How do I achieve that programmatically?

On Thu, 10 Oct 2024, 22:31 Asaf Zorea, @.***> wrote:

The input to noisereduce is a 2-dimensional matrix, so you can input stereo into the algorithm. However, the noise estimation is not shared between the channels.

— Reply to this email directly, view it on GitHub https://github.com/timsainb/noisereduce/issues/114#issuecomment-2406073021, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMSQ2BELAB3I6JZHJXPK4Z3Z23W3DAVCNFSM6AAAAABOOB7U42VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMBWGA3TGMBSGE . You are receiving this because you authored the thread.Message ID: @.***>

[nuniz]

Take a look at the last part (Multichannel noise) of the example notebook - https://github.com/timsainb/noisereduce/blob/master/notebooks/1.0-test-noise-reduction.ipynb

[olawalejuwonm]

Thanks for the tip!

But will the output audio file be stereo as well?

On Fri, Oct 11, 2024 at 9:13 AM Asaf Zorea @.***> wrote:

Take a look at the last part (Multichannel noise) of the example notebook -

https://github.com/timsainb/noisereduce/blob/master/notebooks/1.0-test-noise-reduction.ipynb

— Reply to this email directly, view it on GitHub https://github.com/timsainb/noisereduce/issues/114#issuecomment-2406857877, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMSQ2BGA7MXQ4NRJRW756JLZ26CDVAVCNFSM6AAAAABOOB7U42VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMBWHA2TOOBXG4 . You are receiving this because you authored the thread.Message ID: @.***>

[nuniz]

Yes. The output shape should be identical to the input shape.

[olawalejuwonm]

Ok, thanks.

I noticed this in the codebase:

audio_clip_cafe_2_channel = np.vstack([audio_clip_cafe, audio_clip_cafe])

reduced_noise = nr.reduce_noise(y = audio_clip_cafe_2_channel, sr=rate, n_std_thresh_stationary=1.5,stationary=True)

Do I need to pass the audio to numpy first before using reduced_noise?

On Fri, Oct 11, 2024 at 1:23 PM Asaf Zorea @.***> wrote:

Yes. The output shape should be identical to the input shape.

— Reply to this email directly, view it on GitHub https://github.com/timsainb/noisereduce/issues/114#issuecomment-2407302970, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMSQ2BGUNCK2GDXN63KVOF3Z267MXAVCNFSM6AAAAABOOB7U42VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMBXGMYDEOJXGA . You are receiving this because you authored the thread.Message ID: @.***>

[nuniz]

You can use either numpy or torch.

The simplest usage is:

import soundfile as sf
import noisereduce as nr

# READ AUDIO
data, sr= sf.read("<PATH_TO_AUDIO>")

# REDUCE NOISE
output = nr.reduce_noise(y = data, sr=sr)

# WRITE AUDIO
sf.write("<PATH_TO_AUDIO>", output , sr)

[olawalejuwonm]

What will be the value of

audio_clip_cafe_2_channel ?

On Fri, 11 Oct 2024, 19:33 Asaf Zorea, @.***> wrote:

You can use either numpy or torch.

The simplest usage is:

import soundfile as sf import noisereduce as nr

READ AUDIO

data, sr= sf.read("")

REDUCE NOISE

output = nr.reduce_noise(y = audio_clip_cafe_2_channel, sr=sr)

WRITE AUDIO

sf.write("", output , sr)

— Reply to this email directly, view it on GitHub https://github.com/timsainb/noisereduce/issues/114#issuecomment-2407935887, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMSQ2BATASWAKGRW7WJLMSLZ3AKWHAVCNFSM6AAAAABOOB7U42VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMBXHEZTKOBYG4 . You are receiving this because you authored the thread.Message ID: @.***>

[nuniz]

The audio data

[olawalejuwonm]

I don't get.

Can you 🙏 please share the full code?

On Sat, 12 Oct 2024, 13:02 Asaf Zorea, @.***> wrote:

The audio data

— Reply to this email directly, view it on GitHub https://github.com/timsainb/noisereduce/issues/114#issuecomment-2408538592, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMSQ2BFTL52KBXM55J6M6UDZ3EFTRAVCNFSM6AAAAABOOB7U42VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMBYGUZTQNJZGI . You are receiving this because you authored the thread.Message ID: @.***>

[nuniz]

You can use the code I've written to use noisereduce on stereo-data.

https://github.com/timsainb/noisereduce/issues/114#issuecomment-2407935887