This page demonstrates the parsing process for audio events.
Table of Contents
First, let's import some libraries and prepare the environment for our sample event logs:
>>> import os >>> import tempfile >>> import numpy as np >>> # Prepare temp dirs for storing event files >>> tmpdirs = {}
Before parsing a event file, we need to generate it first. The sample event files are generated by three commonly used event log writers.
PyTorch
We can generate the events by PyTorch:
>>> tmpdirs['torch'] = tempfile.TemporaryDirectory() >>> from torch.utils.tensorboard import SummaryWriter >>> log_dir = tmpdirs['torch'].name >>> rate = 22050 # samples per second >>> T = 3 # sample duration (seconds) >>> f = 440.0 # sound frequency (Hz) >>> t = np.linspace(0, T, T*rate, endpoint=False) >>> x = np.sin(2*np.pi * f * t) >>> x = np.expand_dims(x, axis=0) # single channel >>> writer = SummaryWriter(log_dir) >>> writer.add_audio('my_audio', x, 0, rate) >>> writer.close()
and quickly check the results:
>>> from tbparse import SummaryReader >>> SummaryReader(log_dir, pivot=True).audio.columns Index(['step', 'my_audio'], dtype='object')
TensorFlow2 / Keras
We can generate the events by TensorFlow2 / Keras:
>>> tmpdirs['tensorflow'] = tempfile.TemporaryDirectory() >>> import tensorflow as tf >>> log_dir = tmpdirs['tensorflow'].name >>> rate = 22050 # samples per second >>> T = 3 # sample duration (seconds) >>> f = 440.0 # sound frequency (Hz) >>> t = np.linspace(0, T, T*rate, endpoint=False) >>> x = np.sin(2*np.pi * f * t) >>> x = np.expand_dims(x, axis=1) # single channel >>> x = np.expand_dims(x, axis=0) # batch size = 1 >>> x = x.astype(np.float32) >>> writer = tf.summary.create_file_writer(log_dir) >>> writer.set_as_default() >>> assert tf.summary.audio('my_audio', x, rate, step=0) >>> writer.close()
and quickly check the results:
>>> from tbparse import SummaryReader >>> SummaryReader(log_dir, pivot=True).tensors.columns Index(['step', 'my_audio'], dtype='object')
Warning
The tf.summary.audio function has a parameter max_outputs that limits the maximum number of stored audio instances.
Warning
In the new versions of TensorFlow, the audio method actually stores the events as tensors events inside the event file. Thus, you should perform an extra step with ~tbparse.SummaryReader.tensor_to_audio beforehand if the event file is generated by TensorFlow2. (An example is shown later)
TensorboardX
We can generate the events by TensorboardX:
>>> tmpdirs['tensorboardX'] = tempfile.TemporaryDirectory() >>> from tensorboardX import SummaryWriter >>> log_dir = tmpdirs['tensorboardX'].name >>> rate = 22050 # samples per second >>> T = 3 # sample duration (seconds) >>> f = 440.0 # sound frequency (Hz) >>> t = np.linspace(0, T, T*rate, endpoint=False) >>> x = np.sin(2*np.pi * f * t) >>> x = np.expand_dims(x, axis=1) # single channel >>> writer = SummaryWriter(log_dir) >>> writer.add_audio('my_audio', x, 0, rate) >>> writer.close()
and quickly check the results:
>>> from tbparse import SummaryReader >>> SummaryReader(log_dir, pivot=True).audio.columns Index(['step', 'my_audio'], dtype='object')
In different use cases, we will want to read the event logs in different styles. We further show different configurations of the tbparse.SummaryReader class.
PyTorch
>>> from tbparse import SummaryReader >>> log_dir = tmpdirs['torch'].name >>> # Long Format >>> df = SummaryReader(log_dir, extra_columns={'sample_rate'}).audio >>> df.columns Index(['step', 'tag', 'value', 'sample_rate'], dtype='object') >>> # Wide Format >>> df = SummaryReader(log_dir, pivot=True, extra_columns={'sample_rate'}).audio >>> df.columns Index(['step', 'my_audio', 'my_audio/sample_rate'], dtype='object')
TensorFlow2 / Keras
>>> from tbparse import SummaryReader >>> log_dir = tmpdirs['tensorflow'].name >>> # Long Format >>> df = SummaryReader(log_dir).tensors >>> df.columns Index(['step', 'tag', 'value'], dtype='object') >>> audio_dict_arr = df['value'].apply(SummaryReader.tensor_to_audio) >>> df['value'] = audio_dict_arr.apply(lambda x: x['audio']) >>> df['sample_rate'] = audio_dict_arr.apply(lambda x: x['sample_rate']) >>> df.columns Index(['step', 'tag', 'value', 'sample_rate'], dtype='object') >>> # Wide Format >>> df = SummaryReader(log_dir, pivot=True).tensors >>> df.columns Index(['step', 'my_audio'], dtype='object') >>> audio_dict_arr = df['my_audio'].apply(SummaryReader.tensor_to_audio) >>> df['my_audio'] = audio_dict_arr.apply(lambda x: x['audio']) >>> df['my_audio/sample_rate'] = audio_dict_arr.apply(lambda x: x['sample_rate']) >>> df.columns Index(['step', 'my_audio', 'my_audio/sample_rate'], dtype='object')
TensorboardX
>>> from tbparse import SummaryReader >>> log_dir = tmpdirs['tensorboardX'].name >>> # Long Format >>> df = SummaryReader(log_dir, extra_columns={'sample_rate'}).audio >>> df.columns Index(['step', 'tag', 'value', 'sample_rate'], dtype='object') >>> # Wide Format >>> df = SummaryReader(log_dir, pivot=True, extra_columns={'sample_rate'}).audio >>> df.columns Index(['step', 'my_audio', 'my_audio/sample_rate'], dtype='object')
Warning
When accessing SummaryReader.images, the events stored in each event file are collected internally. The best practice is to store the returned results in a DataFrame as shown in the samples, instead of repeatedly accessing SummaryReader.images.
See the tbparse_extra-columns page for more details.
We further demonstrate some basic techniques for plotting our data.
PyTorch
The data from tensorboard event logs:
import matplotlib.pyplot as plt from tbparse import SummaryReader log_dir = tmpdirs['torch'].name
df = SummaryReader(log_dir, extra_columns={'sample_rate'}).audio x = df.loc[0, 'value'] rate = int(df.loc[0, 'sample_rate'])
T = len(x)//rate t = np.linspace(0, T, T*rate, endpoint=False) plt.plot(t[:200], x[:200])
TensorFlow2 / Keras
The data from tensorboard event logs:
import matplotlib.pyplot as plt from tbparse import SummaryReader log_dir = tmpdirs['tensorflow'].name
df = SummaryReader(log_dir).tensors audio_dict_arr = df['value'].apply(SummaryReader.tensor_to_audio) df['value'] = audio_dict_arr.apply(lambda x: x['audio']) df['sample_rate'] = audio_dict_arr.apply(lambda x: x['sample_rate']) x = df.loc[0, 'value'] rate = int(df.loc[0, 'sample_rate'])
T = len(x)//rate t = np.linspace(0, T, T*rate, endpoint=False) plt.plot(t[:200], x[:200])
TensorboardX
The data from tensorboard event logs:
import matplotlib.pyplot as plt from tbparse import SummaryReader log_dir = tmpdirs['tensorboardX'].name
df = SummaryReader(log_dir, extra_columns={'sample_rate'}).audio x = df.loc[0, 'value'] rate = int(df.loc[0, 'sample_rate'])
T = len(x)//rate t = np.linspace(0, T, T*rate, endpoint=False) plt.plot(t[:200], x[:200])
Target
The ground truth data:
import matplotlib.pyplot as plt
rate = 22050 # samples per second T = 3 # sample duration (seconds) f = 440.0 # sound frequency (Hz) t = np.linspace(0, T, T*rate, endpoint=False) x = np.sin(2*np.pi * f * t) x = np.expand_dims(x, axis=1) # single channel
plt.plot(t[:200], x[:200])