Mixing an audio file with a noise file at any Signal-to-Noise Ratio

The script create_mixed_audio_file.py and create_mixed_audio_file_with_soundfile.py can mix an audio file with a noise file at any Signal-to-Noise Ratio (SNR).

You can listen to the results of mixtures of the clean voice and the noise.

Installation

• Python3.7+
• macOS

Setup

I recommend to install Pipenv for making set-up easy before proceeding:

If you're on macOS, you can install Pipenv easily with Homebrew:

$ brew install pipenv

After installing Pipenv, you run the following command to build a virtualenv and install packages listed in Pipfile.lock.

$ pipenv install

Then, you can run the following command to activate a virtualenv.

$ pipenv shell

File format

To use this code, the format of source waveforms is as follows.

• WAV file
• 1 channel

Additionally, the sampling rate of a clean file and that of a noise file are supposed to be the same.

Usage

There are two files for creating a mixture.

• create_mixed_audio_file.py:

  • Uses wave module
  • Can read wav file with 16-bit PCM only

• create_mixed_audio_file_with_soundfile.py:

  • Uses soundfile library
  • Can read wav file with various encoding types such as 16-bit PCM, 32-bit PCM, 32-bit float, and 64-bit float.

After activating a virtualenv, you can run the files to mix an audio file with a noise file at any signal-to-noise ratio.

Example of create_mixed_audio_file.py:

python create_mixed_audio_file.py --clean_file ~/workspace/audio-SNR/data/16_bit/source_clean/arctic_a0001.wav --noise_file ~/workspace/audio-SNR/data/16_bit/source_noise/ch01.wav --snr 0 --output_mixed_file ~/workspace/audio-SNR/data/16_bit/output_mixed/0dB.wav

Example of create_mixed_audio_file_with_soundfile.py:

python create_mixed_audio_file_with_soundfile.py --clean_file ~/workspace/audio-SNR/data/64_bit/source_clean/arctic_a0001_64bit.wav --noise_file ~/workspace/audio-SNR/data/64_bit/source_noise/ch01_64bit.wav --snr 0 --output_mixed_file ~/workspace/audio-SNR/data/64_bit/output_mixed/0dB.wav

Dataset

I really appreciate the following public datasets.

• Voices - CMU_ARCTIC speech synthesis databases
• Noises - DEMAND: a collection of multi-channel recordings of acoustic noise in diverse environments

Detail

There is a detail about create_mixed_audio_file.py on my post (in Japanese).

https://engineering.linecorp.com/ja/blog/voice-waveform-arbitrary-signal-to-noise-ratio-python/

There is a detail about create_mixed_audio_file.py on my post (in Korean).

https://engineering.linecorp.com/ko/blog/voice-waveform-arbitrary-signal-to-noise-ratio-python/