Athena-signal is an open-source implementation of speech signal processing algorithms. It aims to help reserchers and engineers who want to use speech signal processing algorithms in their own projects. Athena-signal is mainly implemented using C, and called by python.
Currently athena-signal is composed of such modules as Acoustic Echo Cancellation(AEC), High Pass Filter(HPF), Direction Of Arrival(DOA), Minimum Variance Distortionless Response(MVDR) beamformer, Generalized Sidelobe Canceller(GSC), Voice Activity Detection(VAD), Noise Supression(NS), and Automatic Gain Control(AGC).
- AEC: The core of the AEC algorithm includes time delay estimation, linear echo cancellation, double-talk detection, echo return loss estimation and residual echo suppression.
- HPF: High-pass filtering is implemented using cascaded-iir-filter. The cut-off frequency is 200Hz in this program. You can rewrite the iir filter coefficients and gains, with the help of filter design toolbox in MATLAB, to generate high-pass filter with cut-off frequency you set.
- DOA: Capon algorithm is used to get the direction of the sound source. The main function of the Capon algorithm is the Capon beamformer, also called MVDR. The Capon spectrum is estimated by using Rxx matrix and steering vector in frequency domain.
- MVDR: This is a Minimum Variance Distortionless Response beamformer. You can set the steering vector(loc_phi) with the help of DOA estimation, which indicates the distortionless response direction. Rnn matrix is estimated using MCRA noise estimation method. Microphone array could be any shape as long as you set the coordinates of each microphones(mic_coord) beforehand. In the future edition, the steering vector will be estimated by DOA estimation. You can set the steering vector by your own DOA estimation method of course.
- GSC: This is a Generalized Sidelobe Canceller beamformer, It is composed of Fixed Beamformer(FBF), Adaptive Blocking Matrix(ABM) and the Adaptive Interference Canceller(AIC) modules.
- VAD: Voice Activity Detection(VAD) function outputs the current frame speech state based on the result of the double-talk detection.
- NS: Noise reduction algorithm is based on MCRA noise estimation method. Details can be found in "Noise Estimation by Minima Controlled Recursive Averaging for Robust Speech Enhancement" and "Noise Spectrum Estimation in Adverse Environments: Improved Minima Controlled Recursive Averaging"
- AGC: Automatic Gain Control(AGC) determines the gain factor based on the current frame signal level and the target level so that the gain of the signal is kept within a reasonable range.
Each modules has a separate switch that controls the operating status of the module. In the code, modules are turned off by default except AEC and NS.
- You need to set the switches of each module to determine whether it is enabled or not. Enter the number of mic and reference, and the coordinates of each microphone. When the switch is set to 1, it is in the running state, and when it is 0, the module will be skipped. Generally you have to manually set the number of microphones and reference channels, and the coordinates for each microphone. The coordinates MUST be set when MVDR, GSC or DOA module is enabled.
- You can set the length of the read and write data array_frm_len, and it is 128 by default, ptr_input_data and ptr_ref_data store the microphone signal data and reference signal data, respectively. The multi-channel microphone signals are stored in the form of parallel input, that is, the data of each channel is sequentially stored in ptr_input_data.
- Since MVDR requires the angle of incidence of the sound source, we set it to 90 by default. When the DOA module is enabled, the steering vector will be estimated by DOA estimation. MVDR supports ANY array setups, including circular array and linear array, as long as you set the coordinates of microphones mic_coord beforehand.
- Python3.x
- swig
- numpy
- setuptools
- Linux
- MacOS
- Windows
swig -python athena_signal/dios_signal.i
python setup.py bdist_wheel sdist
#For Linux or MacOS
pip install --ignore-installed dist/athena_signal-0.1.0-*.whl
#For Windows
for /r dist %%i in (athena_signal-*.whl) do pip install --ignore-installed %%i
python examples/athena_signal_test.py
config(dictionary):
--add_AEC : If 1, do AEC on signal.
(int, default = 1)
--add_NS : If 1, do NS on signal.
(int, default = 1)
--add_AGC : If 1, do AGC on signal.
(int, default = 0)
--add_HPF : If 1, do HPF on signal.
(int, default = 0)
--add_BF : If 1, do MVDR on signal.
(int, default = 0)
: If 2, do GSC on signal.
(int, default = 0)
--add_DOA : If 1, do DOA on signal.
(int, default = 0)
--mic_num : Number of microphones.
(int, default = 1)
--ref_num : Number of reference channel.
(int, default = 1)
mic_coord(array/list):
The coordinates of each microphone of the microphone array
using in MVDR. (A float array/list of size [mic_num, 3] containing
three-dimensional coordinates of every microphone.)
from athena_signal.dios_ssp_api import athena_signal_process
#Test AEC
input_file = ["examples/0841-0875_env7_sit1_male_in.pcm"]
ref_file = ["examples/0841-0875_env7_sit1_male_ref.pcm"]
out_file = ["examples/0841-0875_env7_sit1_male_out.pcm"]
config = {'add_AEC': 1, 'add_BF': 0}
athena_signal_process(input_file, out_file, ref_file, config)
# Test BF
input_file = ["examples/m0f60_5cm_1_mix.pcm",
"examples/m0f60_5cm_2_mix.pcm",
"examples/m0f60_5cm_3_mix.pcm",
"examples/m0f60_5cm_4_mix.pcm",
"examples/m0f60_5cm_5_mix.pcm",
"examples/m0f60_5cm_6_mix.pcm"]
out_file = ["examples/m0f60_5cm_mvdr_out.pcm"]
config = {'add_AEC': 0, 'add_BF': 1, 'add_DOA': 1, 'mic_num': 6}
mic_coord = [[0.05, 0.0, 0.0],
[0.025, 0.0433, 0.0],
[-0.025, 0.0433, 0.0],
[-0.05, 0.0, 0.0],
[-0.025, -0.0433, 0.0],
[0.025, -0.0433, 0.0]]
athena_signal_process(input_file, out_file, config=config, mic_coord=mic_coord)
Any contribution is welcome. All issues and pull requests are highly appreciated. If you have any questions when using athena-signal, or any ideas to make it better, please feel free to contact us.
Athena-signal is built with the help of some open-source repos such as WebRTC, speex, etc.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent