diff --git a/niimpy/preprocessing/audio.py b/niimpy/preprocessing/audio.py index d6f5198d..a4209593 100755 --- a/niimpy/preprocessing/audio.py +++ b/niimpy/preprocessing/audio.py @@ -125,7 +125,7 @@ def audio_count_loud(df_u, config=None): df_u[col_name] = pd.to_numeric(df_u[col_name]) if len(df_u)>0: - df_s = df_u[df_u[col_name]>70] #check if environment was noisy + df_u = df_u[df_u[col_name]>70] #check if environment was noisy result = util.group_data(df_u)[col_name].resample(**config["resample_args"]).count() result = result.to_frame(name='audio_count_loud') result = util.reset_groups(result)