with retrieved shape (4, 32) does not match shape=[5, 32] dtype=dtype('complex64')

[aleksandra-bebe]

with retrieved shape (4, 32) does not match shape=[5, 32] dtype=dtype('complex64')

[jmcasebeer]

Hello, and thanks for the interest. Could you please share the exact code you ran, and describe what setup you did before that?

[aleksandra-bebe]

@jmcasebeer I downloaded pre-trained models from the link and provided the path to the AEC model. Additionally, I downloaded the AEC challenge dataset and specified its path in the config.py script." and on this line of code pred = system.infer({"signals": d_input, "metadata": {}}, fit_infer=fit_infer)[0] give me issue
This is the code that I ran.
import os
from aec_eval import get_system_ckpt
import numpy as np
import librosa
import soundfile as sf
import aec

ckpt_dir = r"C:\Users\pc\Desktop\AI-Beamformers\meta-af\MetaAF\v0.1.0_models\aec"
name = "aec_16_dt_c"
date = "2022_04_10_15_57_12"
epoch = 230

ckpt_loc = os.path.join(ckpt_dir, name, date)

system, kwargs, outer_learnable = get_system_ckpt(
ckpt_loc,
epoch,
model_type="egru",
system_len=None,
)
fit_infer = system.make_fit_infer(outer_learnable=outer_learnable)
fs = 16000

out_dir = "metaAF_res"
os.makedirs(out_dir, exist_ok=True)

u, _ = librosa.load(r"C:\Users\pc\Desktop\AI-Beamformers\meta-af\MetaAF\zoo\aec\u.wav", sr=fs)
d, _ = librosa.load(r"C:\Users\pc\Desktop\AI-Beamformers\meta-af\MetaAF\zoo\aec\d.wav", sr=fs)
s, _ = librosa.load(r"C:\Users\pc\Desktop\AI-Beamformers\meta-af\MetaAF\zoo\aec\s.wav", sr=fs)
e = d - s
d_mp3_input = {"u": u[None, :, None], "d": d[None, :, None],
"s": s[None, :, None], "e": e[None, :, None]}

pred_mp3 = system.infer({"signals": d_mp3_input, "metadata": {}}, fit_infer=fit_infer)[
0
]
pred_mp3 = np.array(pred_mp3[0, :, 0])

dset = aec.MSFTAECDataset_RIR(mode='test', double_talk=True, random_roll=True, scene_change=False)
data = dset[0]
u, d, e, s = (
data["signals"]["u"],
data["signals"]["d"],
data["signals"]["e"],
data["signals"]["s"],
)
d_input = {"u": u[None], "d": d[None], "s": s[None], "e": e[None]}

pred = system.infer({"signals": d_input, "metadata": {}}, fit_infer=fit_infer)[
0
]
pred = np.array(pred[0, :, 0])

sf.write(os.path.join(out_dir, f"_out.wav"), pred, fs)

[aleksandra-bebe]

@jmcasebeer
Hello! I hope you had a wonderful holidays. Do you have any solution regarding my issue?

[jmcasebeer]

Could you please try using the most recent model weights? It looks like you're using the 0.1.0 weights with the 1.0.1 code. You can get the 1.0.1 weights here. Then, update the checkpoint loading to use a more recent checkpoint.

For example:

ckpt_dir = "v1.0.1_models/aec/"
name = "meta_aec_16_combo_rl_4_1024_512_r2"
date = "2022_10_19_23_43_22"
epoch = 110

ckpt_loc = os.path.join(ckpt_dir, name, date)

system, kwargs, outer_learnable = get_system_ckpt(
    ckpt_loc,
    epoch,
)
fit_infer = system.make_fit_infer(outer_learnable=outer_learnable)
fs = 16000

[aleksandra-bebe]

@jmcasebeer Thank you very much, now it work !