Merge branch 'ci/no-padding-for-wavdiff' into 'main' (28a985c2) · Commits · IVAS Codec Public Collaboration / IVAS Codec

scripts/pyaudio3dtools/audioarray.py

+63 −18

Original line number	Diff line number	Diff line
		@@ -39,7 +39,8 @@ import shutil
		import subprocess
		import tempfile
		from pathlib import Path
		from typing import Callable, Iterable, Optional, Tuple
		from enum import Enum
		from typing import Callable, Iterable, Optional, Tuple, Union

		import numpy as np
		import scipy.io.wavfile as wavfile
		@@ -228,6 +229,12 @@ def cut(x: np.ndarray, limits: Tuple[int, int]) -> np.ndarray:
		return y


		class HandleDifferingLengths(str, Enum):
		FAIL = "fail"
		PAD = "pad"
		CUT = "cut"


		def compare(
		ref: np.ndarray,
		test: np.ndarray,
		@@ -239,8 +246,9 @@ def compare(
		ssnr_thresh_high: float = np.inf,
		apply_thresholds_to_ref_only: bool = False,
		test_start_offset_ms: int = 0,
		ref_jbm_tf: Optional[Path] = None,
		test_jbm_tf: Optional[Path] = None,
		ref_jbm_tf: Optional[Union[Path, str]] = None,
		test_jbm_tf: Optional[Union[Path, str]] = None,
		handle_differing_lengths: HandleDifferingLengths = "fail",
		) -> dict:
		"""Compare two audio arrays

		@@ -271,6 +279,17 @@ def compare(
		test_start_offset_ms: (non-negative) int
		offset in miliseconds for test signal. If > 0, the corresponding number of samples
		will be removed from the test array like so: test = test[sample_offset:, :].
		ref_jbm_tf: str\|Path
		tracefile for ref signal for wav-diff based MLD comparison of JBM output
		test_jbm_tf: str\|Path
		tracefile for test signal for wav-diff based MLD comparison of JBM output
		handle_differing_lengths: one of "fail", "pad", "cut"
		how to handle differing lengths in the input signals
		"fail" - raise error
		"pad" - pad shorter file with zeros
		"cut" - cut longer file to length of shorter one

		Note that external tools such as wav-diff (for mld) always use the unmodified files

		Returns
		-------
		@@ -286,11 +305,38 @@ def compare(
		test = test[test_start_offset_samples:, :]

		framesize = fs // 50
		if ref.shape[0] != test.shape[0]:

		lengths_differ = ref.shape[0] != test.shape[0]

		test_orig = test.copy()
		ref_orig = ref.copy()

		if lengths_differ:
		if handle_differing_lengths == "fail":
		raise RuntimeError(
		f"Input signals have different lengths: ref - {ref.shape[0]}, test - {test.shape[0]}"
		)
		elif handle_differing_lengths == "cut":
		min_len = min(ref.shape[0], test.shape[0])
		diff = abs(test[:min_len, :] - ref[:min_len, :])
		else:
		ref = ref[:min_len, :]
		test = test[:min_len, :]
		elif handle_differing_lengths == "pad":
		max_len = max(ref.shape[0], test.shape[0])
		ref = np.pad(
		ref,
		((0, max_len - ref.shape[0]), (0, 0)),
		mode="constant",
		constant_values=0,
		)
		test = np.pad(
		test,
		((0, max_len - test.shape[0]), (0, 0)),
		mode="constant",
		constant_values=0,
		)

		diff = abs(test - ref)

		max_diff = int(diff.max())
		result = {
		"bitexact": True,
		@@ -363,6 +409,8 @@ def compare(
		if get_mld:

		def parse_wav_diff(proc: subprocess.CompletedProcess) -> float:
		if proc.returncode:
		raise ChildProcessError(f"{proc.stderr}\n{proc.stdout}")
		line = proc.stdout.splitlines()[-1].strip()
		start = line.find(">") + 1
		stop = line.rfind("<")
		@@ -370,11 +418,6 @@ def compare(

		return mld

		# TODO probably needs a fix to show up in pytest
		if proc.returncode:
		print(f"{proc.stderr}\n{proc.stdout}")
		return mld_max

		mld_max = 0
		toolsdir = Path(__file__).parent.parent.joinpath("tools")

		@@ -397,18 +440,17 @@ def compare(
		tmpfile_ref = Path(tmpdir).joinpath("ref.wav")
		tmpfile_test = Path(tmpdir).joinpath("test.wav")


		### need to resample to 48kHz for MLD computation to be correct
		if fs != 48000:
		ref_tmp = np.clip(
		resample(ref.astype(float), fs, 48000), -32768, 32767
		resample(ref_orig.astype(float), fs, 48000), -32768, 32767
		)
		test_tmp = np.clip(
		resample(test.astype(float), fs, 48000), -32768, 32767
		resample(test_orig.astype(float), fs, 48000), -32768, 32767
		)
		else:
		ref_tmp = ref.copy()
		test_tmp = test.copy()
		ref_tmp = ref_orig.copy()
		test_tmp = test_orig.copy()

		wavfile.write(str(tmpfile_ref), 48000, ref_tmp.astype(np.int16))
		wavfile.write(str(tmpfile_test), 48000, test_tmp.astype(np.int16))
		@@ -416,6 +458,9 @@ def compare(
		cmd = [
		str(wdiff),
		"--print-ctest-measurement",
		# wav-diff return code is 1 if differences are found which
		# would cause parse_wav_diff to raise an Exception on these cases
		"--no-fail",
		str(tmpfile_ref),
		str(tmpfile_test),
		]

tests/cmp_pcm.py

+24 −10

Original line number	Diff line number	Diff line
		@@ -15,6 +15,7 @@ sys.path.append(os.path.join(os.path.dirname(THIS_PATH), "../scripts"))
		import numpy as np
		import pyaudio3dtools
		import pyivastest

		# Hack to resolve import when using from command line or from within scripts.
		try:
		from .constants import ODG_PATTERN_PQEVALAUDIO
		@@ -38,7 +39,7 @@ def cmp_pcm(
		odg_ref=None,
		ref_jbm_tf: Optional[Path] = None,
		cut_jbm_tf: Optional[Path] = None,
		) -> (int, str):
		) -> tuple[int, str]:
		"""
		Compare 2 PCM files for bitexactness
		"""
		@@ -77,11 +78,9 @@ def cmp_pcm(
		reason = "FAIL: Number of channels differ."
		return 1, reason

		handle_differing_lengths = "fail"
		if allow_differing_lengths:
		# to allow for MLD comparison, pad shorter file
		max_len = max(s1.shape[0], s2.shape[0])
		s1 = np.pad(s1,((0,max_len - s1.shape[0]),(0,0)),mode='constant',constant_values=0)
		s2 = np.pad(s2,((0,max_len - s2.shape[0]),(0,0)),mode='constant',constant_values=0)
		handle_differing_lengths = "pad"
		elif s1.shape != s2.shape:
		print(
		f"file size in samples: file 1 = {s1.shape[0]},",
		@@ -103,6 +102,7 @@ def cmp_pcm(
		ssnr_thresh_low=-50,
		ref_jbm_tf=ref_jbm_tf,
		test_jbm_tf=cut_jbm_tf,
		handle_differing_lengths=handle_differing_lengths,
		)

		output_differs = 0
		@@ -149,13 +149,23 @@ def cmp_pcm(
		32767,
		).astype(np.int16)

		pqeval_output = pqevalaudio_wrapper(odg_files[odg_input], odg_files[odg_ref], 48000)
		pqeval_output = pqevalaudio_wrapper(
		odg_files[odg_input], odg_files[odg_ref], 48000
		)
		match_odg = re.search(ODG_PATTERN_PQEVALAUDIO, pqeval_output)
		try:
		odg_ref = float(match_odg.groups()[0])
		except AttributeError:
		raise OdgParsingFailed("Could not get Odg for ref signal")

		pqeval_output = pqevalaudio_wrapper(odg_files[odg_input], odg_files[odg_test], 48000)
		pqeval_output = pqevalaudio_wrapper(
		odg_files[odg_input], odg_files[odg_test], 48000
		)
		match_odg = re.search(ODG_PATTERN_PQEVALAUDIO, pqeval_output)
		try:
		odg_test = float(match_odg.groups()[0])
		except AttributeError:
		raise OdgParsingFailed("Could not get Odg for test signal")

		odg = odg_test - odg_ref # Todo: store both rather than difference?

		@@ -166,6 +176,10 @@ def cmp_pcm(
		return output_differs, reason


		class OdgParsingFailed(Exception):
		pass


		def pqevalaudio_wrapper(
		ref_sig: np.ndarray,
		eval_sig: np.ndarray,