Commit f141f817 authored by PLAINSI's avatar PLAINSI
Browse files

Next draft

parent 42bbe28d

item_gen_configs/FOA_CONFIG.yml

+6 −6
Original line number Diff line number Diff line
@@ -56,9 +56,9 @@ scenes: 
        IR: ["AmbiX_036x033y_0.01m.wav", "AmbiX_-70x035y_0.01m.wav"]

        overlap: 1.0

        

    a2: 

        name: "G6S2.wav"

        description: "Two speakers sitting in a car."

        source: ["fa1.wav", "ma1.wav"]

        IR: ["AmbiX_036x033y_0.01m.wav", "AmbiX_-70x035y_0.01m.wav"]

        overlap: 1.0
 
    #a2: 

    #    name: "G6S2.wav"

    #    description: "Two speakers sitting in a car."

    #    source: ["fa1.wav", "ma1.wav"]

    #    IR: ["AmbiX_036x033y_0.01m.wav", "AmbiX_-70x035y_0.01m.wav"]

    #    overlap: 1.0

ivas_processing_scripts/audiotools/wrappers/reverb.py

+0 −66
Original line number Diff line number Diff line
@@ -184,69 +184,3 @@ def reverb_stereo( 
    y.audio = np.column_stack([y_left.audio, y_right.audio])



    return y
 


def reverb_foa(

    input: Audio,

    foa_IR: Audio,

    align: Optional[float] = None,

) -> Audio:

    """

    Wrapper for the ITU-T reverb binary to convolve mono audio signal with an FOA impulse response



    Parameters

    ----------

    input: Audio

        Input audio signal

    IR: Audio

        Impulse response

    align: float

         multiplicative factor to apply to the reverberated sound in order to align its energy level with the second file



    Returns

    -------

    output: Audio

        Convolved audio signal with FOA IR

    """



    # convert to float32

    foa_IR.audio = np.float32(foa_IR.audio)



    # separate into each channel

    IR_w = copy(foa_IR)

    IR_w.name = "MONO"

    IR_w.num_channels = 1

    IR_w.audio = np.reshape(foa_IR.audio[:, 0], (-1, 1))



    IR_x = copy(foa_IR)

    IR_x.name = "MONO"

    IR_x.num_channels = 1

    IR_x.audio = np.reshape(foa_IR.audio[:, 1], (-1, 1))



    IR_y = copy(foa_IR)

    IR_y.name = "MONO"

    IR_y.num_channels = 1

    IR_y.audio = np.reshape(foa_IR.audio[:, 2], (-1, 1))



    IR_z = copy(foa_IR)

    IR_z.name = "MONO"

    IR_z.num_channels = 1

    IR_z.audio = np.reshape(foa_IR.audio[:, 3], (-1, 1))



    # calculate the scaling (multiplicative) factor such that the maximum gain of the IR filter across all frequencies is 0dB

    if align is None:

        H = fft(foa_IR.audio, axis=0)

        align = 1.0 / np.max(np.abs(H))



    # convolve mono input with left and right IR

    y_w = reverb(input, IR_w, align=align)

    y_x = reverb(input, IR_x, align=align)

    y_y = reverb(input, IR_y, align=align)

    y_z = reverb(input, IR_z, align=align)



    # combine into foa output

    y = copy(input)

    y.name = "FOA"

    y.num_channels = 4

    y.audio = np.column_stack([y_w.audio, y_x.audio, y_y.audio, y_z.audio])



    return y

ivas_processing_scripts/generation/process_foa_items.py

+112 −5
Original line number Diff line number Diff line
@@ -32,13 +32,18 @@ 


import logging

import os

from copy import copy



from math import floor

from typing import Optional



import numpy as np

import scipy.signal as ssg

from scipy.fft import fft



from ivas_processing_scripts.audiotools.audio import Audio

from ivas_processing_scripts.audiotools import audio, audiofile

from ivas_processing_scripts.audiotools.wrappers.bs1770 import get_loudness

from ivas_processing_scripts.audiotools.wrappers.reverb import reverb_foa

from ivas_processing_scripts.generation import config



SEED_RANDOM_NOISE = 0
@@ -49,6 +54,108 @@ def csv_formatdata(data): 
    for row in data:

        yield ["%0.2f" % v for v in row]



def filter_one(

    input: Audio,

    IR: Audio,

    align: Optional[float] = None,

) -> Audio:

    """

    Parameters

    ----------

    input: Audio

        Input audio signal

    IR: Audio

        Impulse response

    align: float

         multiplicative factor to apply to the reverberated sound in order to align its energy level with a second filePath to the output file



    Returns

    -------

    output: Audio

        Convolved audio signal with IR

    """



    # resample IR to input signal

    tmp_IR = copy(IR)

    if input.fs != IR.fs:

        tmp_IR.audio = ssg.resample_poly(IR.audio, input.fs, IR.fs) # resample_itu(tmp_input, IR.fs)

        tmp_IR.fs = input.fs



    # down-scale IR to prevent saturation

    # max_value = np.max(np.abs(IR.audio))

    # if max_value > 1.0:

    # IR.audio = IR.audio / max_value



    tmp_IR.audio = tmp_IR.audio * align

    output = copy(input)

    output.audio[0] = ssg.lfilter(tmp_IR.audio[0],1,input.audio[0])



    return output



def filter_foa(

    input: Audio,

    foa_IR: Audio,

    align: Optional[float] = None,

) -> Audio:

    """

    Parameters

    ----------

    input: Audio

        Input audio signal

    IR: Audio

        Impulse response

    align: float

         multiplicative factor to apply to the reverberated sound in order to align its energy level with the second file



    Returns

    -------

    output: Audio

        Convolved audio signal with FOA IR

    """



    # convert to float32

    foa_IR.audio = np.float32(foa_IR.audio)



    # separate into each channel

    IR_w = copy(foa_IR)

    IR_w.name = "MONO"

    IR_w.num_channels = 1

    IR_w.audio = np.reshape(foa_IR.audio[:, 0], (-1, 1))



    IR_x = copy(foa_IR)

    IR_x.name = "MONO"

    IR_x.num_channels = 1

    IR_x.audio = np.reshape(foa_IR.audio[:, 1], (-1, 1))



    IR_y = copy(foa_IR)

    IR_y.name = "MONO"

    IR_y.num_channels = 1

    IR_y.audio = np.reshape(foa_IR.audio[:, 2], (-1, 1))



    IR_z = copy(foa_IR)

    IR_z.name = "MONO"

    IR_z.num_channels = 1

    IR_z.audio = np.reshape(foa_IR.audio[:, 3], (-1, 1))



    # calculate the scaling (multiplicative) factor such that the maximum gain of the IR filter across all frequencies is 0dB

    if align is None:

        H = fft(foa_IR.audio, axis=0)

        align = 1.0 / np.max(np.abs(H))



    # convolve mono input with left and right IR

    y_w = filter_one(input, IR_w, align=align)

    y_x = filter_one(input, IR_x, align=align)

    y_y = filter_one(input, IR_y, align=align)

    y_z = filter_one(input, IR_z, align=align)



    # combine into foa output

    y = copy(input)

    y.name = "FOA"

    y.num_channels = 4

    y.audio = np.column_stack([y_w.audio, y_x.audio, y_y.audio, y_z.audio])



    return y





def generate_foa_items(

    cfg: config.TestConfig,
@@ -96,7 +203,7 @@ def generate_foa_items( 
        N_sources = len(np.atleast_1d(scene["source"]))



        # read the IR (check if foa or two mono files were provided)

        source_IR = np.atleast_1d(scene["IR"])

        # source_IR = np.atleast_1d(scene["IR"])



        # read the overlap length

        if "overlap" in scene.keys():
@@ -111,7 +218,7 @@ def generate_foa_items( 
            print("source file: {}".format(source_file))

            IR_file = np.atleast_1d(scene["IR"])[i]



            logger.info(f"Convolving {source_file} with {source_IR}")

            logger.info(f"Convolving {source_file} with {IR_file}")



            # read source file

            x = audio.fromfile(
@@ -124,10 +231,10 @@ def generate_foa_items( 
            )



            # convolve with FOA IR

            x = reverb_foa(x, IR)

            x = filter_foa(x, IR)



            # adjust the level of the foa signal

            _, scale_factor, _ = get_loudness(x, cfg.loudness, "FOA")

            _, scale_factor, _ = get_loudness(x, cfg.loudness, "BINAURAL")

            x.audio *= scale_factor



            # shift the second (and all other) source files (positive shift creates overlap, negative shift creates a gap)