complement reverb tool with HOA3 format, corrections and fixes (f5121821) · Commits · IVAS Codec Public Collaboration / IVAS Processing Scripts

ivas_processing_scripts/audiotools/wrappers/reverb.py

+87 −18

Original line number	Diff line number	Diff line
		@@ -38,6 +38,7 @@ from typing import Optional
		import numpy as np
		from scipy.fft import fft

		from ivas_processing_scripts.audiotools import audio
		from ivas_processing_scripts.audiotools.audio import Audio
		from ivas_processing_scripts.audiotools.audiofile import read, write
		from ivas_processing_scripts.audiotools.wrappers.filter import resample_itu
		@@ -128,7 +129,9 @@ def reverb(
		output = copy(tmp_input)
		output.audio, _ = read(tmp_output_file, nchannels=1, fs=tmp_input.fs)

		# reverse the resampling
		# remove trailing part (to ensure that the length of the output is the same as the input)
		output.audio = output.audio[:-(IR.audio.shape[0]-1), :]

		if old_fs:
		output.audio = resample_itu(output, old_fs)
		output.fs = old_fs
		@@ -183,9 +186,8 @@ def reverb_stereo(
		y_right = reverb(input, IR_right, align=align)

		# combine into stereo output
		y = copy(input)
		y.name = "STEREO"
		y.num_channels = 2
		y = audio.fromtype('STEREO')
		y.fs = input.fs
		y.audio = np.column_stack([y_left.audio, y_right.audio])

		return y
		@@ -197,14 +199,14 @@ def reverb_foa(
		align: Optional[float] = None,
		) -> Audio:
		"""
		Wrapper for the ITU-T reverb binary to convolve mono audio signal with an FOA impulse response
		Convolve mono audio signal with an FOA impulse response

		Parameters
		----------
		input: Audio
		Input audio signal
		IR: Audio
		Impulse response
		foa_IR: Audio
		FOA impulse response
		align: float
		multiplicative factor to apply to the reverberated sound in order to align its energy level with the second file

		@@ -249,10 +251,9 @@ def reverb_foa(
		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
		# combine into FOA output
		y = audio.fromtype('FOA')
		y.fs = input.fs
		y.audio = np.column_stack([y_w.audio, y_x.audio, y_y.audio, y_z.audio])

		return y
		@@ -264,14 +265,14 @@ def reverb_hoa2(
		align: Optional[float] = None,
		) -> Audio:
		"""
		Wrapper for the ITU-T reverb binary to convolve mono audio signal with an HOA2 impulse response
		Convolve mono audio signal with an HOA2 impulse response

		Parameters
		----------
		input: Audio
		Input audio signal
		IR: Audio
		Impulse response
		hoa2_IR: Audio
		HOA2 impulse response
		align: float
		multiplicative factor to apply to the reverberated sound in order to align its energy level with the second file

		@@ -301,10 +302,71 @@ def reverb_hoa2(
		# convolve mono input with channel IR
		ych.append(reverb(input, IR, align=align))

		# combine into hoa2 output
		y = copy(input)
		y.name = "HOA2"
		y.num_channels = numchannels
		# combine into HOA2 output
		y = audio.fromtype('HOA2')
		y.fs = input.fs
		y.audio = np.column_stack(
		[
		ych[0].audio,
		ych[1].audio,
		ych[2].audio,
		ych[3].audio,
		ych[4].audio,
		ych[5].audio,
		ych[6].audio,
		ych[7].audio,
		ych[8].audio,
		]
		)

		return y

		def reverb_hoa3(
		input: Audio,
		hoa3_IR: Audio,
		align: Optional[float] = None,
		) -> Audio:
		"""
		Convolve mono audio signal with an HOA3 impulse response

		Parameters
		----------
		input: Audio
		Input audio signal
		hoa3_IR: Audio
		HOA3 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 HOA3 IR
		"""

		# convert to float32
		hoa3_IR.audio = np.float32(hoa3_IR.audio)

		numchannels = 16 # HOA3 by definition

		# 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(hoa3_IR.audio, axis=0)
		align = 1.0 / np.max(np.abs(H))

		IR = copy(hoa3_IR)
		IR.name = "MONO"
		IR.num_channels = 1
		ych = []
		for i in range(numchannels):
		# separate IR into each channel
		IR.audio = np.reshape(hoa3_IR.audio[:, i], (-1, 1))
		# convolve mono input with channel IR
		ych.append(reverb(input, IR, align=align))

		# combine into HOA3 output
		y = audio.fromtype('HOA3')
		y.fs = input.fs
		y.audio = np.column_stack(
		[
		ych[0].audio,
		@@ -316,6 +378,13 @@ def reverb_hoa2(
		ych[6].audio,
		ych[7].audio,
		ych[8].audio,
		ych[9].audio,
		ych[10].audio,
		ych[11].audio,
		ych[12].audio,
		ych[13].audio,
		ych[14].audio,
		ych[15].audio,
		]
		)