Latencies for Marti v0

# generate data about the times for each part of the operation # user times is the amount of time the user spends talking # listen times is the amount of time the program spends in listening mode # buffer times is the difference between user times and listen times, which is the amount of time that the program buffers to ensure that the user is done speaking # recognize times is the amount of time that the program takes to turn the user's speech into text # think times is the amount of time that the program takes to return the text output for the user # generate times is the amount of time that the program takes to turn text into speech # read times is the amount of time that the program takes to read the generated audio, not including time playing the audio user_times = [3.53, 1.81, 5.24, 3.14, 3.53, 4.44, 3.49, 3.85, 1.48, 1.86, 2.64, 3.76, 1.86, 2.66, 4.49, 3.47, 3.81, 3.21, 4.50, 3.02, 3.06] listen_times = [6.5022499561309814, 4.427386283874512, 4.326539993286133, 5.218196153640747, 5.100186109542847, 6.614777088165283, 5.056102275848389, 3.220916986465454, 3.147758960723877, 3.7746741771698, 3.650014877319336, 6.38067889213562, 4.603470087051392, 4.430200815200806, 5.703462839126587, 4.781328916549683, 5.39479398727417, 4.716341972351074, 5.652783155441284, 4.413949728012085, 4.455374240875244] buffer_times = [listen_time - user_time for listen_time, user_time in zip(listen_times, user_times)] recognize_times = [1.6909189224243164, 1.0840249061584473, 1.5180540084838867, 2.0668721199035645, 1.692863941192627, 2.8224809169769287, 2.4820170402526855, 1.0097508430480957, 1.0307588577270508, 1.2826569080352783, 1.457226037979126, 1.8728969097137451, 1.4664230346679688, 1.7909152507781982, 2.25940203666687, 2.1372830867767334, 2.610731601715088, 1.1604321002960205, 2.0273590087890625, 2.074542760848999, 1.619128942489624] think_times = [2.1137890815734863, 1.4204962253570557, 1.5143821239471436, 2.2850210666656494, 1.4985020160675049, 1.5348238945007324, 2.040102005004883, 1.7425520420074463, 0.9219210147857666, 1.1603200435638428, 3.7914979457855225, 1.6758239269256592, 1.6205110549926758, 1.842151165008545, 1.8236520290374756, 2.3010358810424805, 1.7801809310913086, 1.4898388385772705, 2.7025678157806396, 1.1539373397827148, 1.0625710487365723] generate_times = [0.956312894821167, 0.49080395698547363, 0.3302750587463379, 0.4471161365509033, 0.3730340003967285, 0.43520092964172363, 0.48029184341430664, 0.4222593307495117, 0.3813629150390625, 0.3173530101776123, 0.5469510555267334, 0.5209412574768066, 0.3671531677246094, 0.4146718978881836, 0.32427191734313965, 1.306190013885498, 0.6348428726196289, 0.7616047859191895, 0.47567319869995117, 0.3313148021697998, 0.35553503036499023] read_times = [0.3130319118499756, 0.14511704444885254, 0.1421198844909668, 0.14771318435668945, 0.14586901664733887, 0.14693403244018555, 0.14595508575439453, 0.1464698314666748, 0.1440601348876953, 0.1438748836517334, 0.14755892753601074, 0.15033888816833496, 0.14429998397827148, 0.14671683311462402, 0.13856887817382812, 0.1457977294921875, 0.14838719367980957, 0.14347600936889648, 0.15015602111816406, 0.14775514602661133, 0.14155888557434082]

#convert the lists into a single pandas df import pandas as pd times_df = pd.DataFrame( {'User Times': user_times, 'Buffer Times': buffer_times, 'Recognize Times': recognize_times, 'Think Times': think_times, 'Generate Times': generate_times, 'Read Times': read_times })

# visualize the latencies of all modules in a single graph of box plots import seaborn as sns import matplotlib.pyplot as plt melted_times = pd.melt(times_df) melted_times.set_axis(["Program Components", "Latencies"], axis = 1, inplace = True) # The box plot all_latencies = sns.boxplot(data = melted_times, x = "Program Components", y = "Latencies") # Show the plot sns.set(rc={'figure.figsize':(15, 10)}) all_latencies.get_figure();

## visualize latencies of only the listening module in a single box plot listen_melted_times = pd.melt(times_df[['User Times', 'Buffer Times', 'Recognize Times']]) listen_melted_times.set_axis(["Program Components", "Latencies"], axis = 1, inplace = True) # The box plot listen_latencies = sns.boxplot(data = listen_melted_times, x = "Program Components", y = "Latencies") # Show the plot sns.set(rc={'figure.figsize':(15, 10)}) listen_latencies.get_figure();

# visualize the latency of thinking with a KDE, with the mean plotted sns.displot(times_df, x="Think Times", kind="kde", aspect = 2) plt.axvline(x=times_df['Think Times'].mean(), color='red')

#visualize the latency of the components of the speak module with violin plots # visualize latencies of only the listening module in a single box plot speak_melted_times = pd.melt(times_df[['Generate Times', 'Read Times']]) speak_melted_times.set_axis(["Program Components", "Latencies"], axis = 1, inplace = True) # The box plot speak_latencies = sns.violinplot(data = speak_melted_times, x = "Program Components", y = "Latencies") # Show the plot sns.set(rc={'figure.figsize':(15, 10)}) speak_latencies.get_figure();

# generate a correlation matrix to test correlations between different latencies corr = times_df.corr() # Set up the matplotlib figure f, ax = plt.subplots(figsize=(7, 9)) ax.xaxis.tick_top() # Generate a custom diverging colormap cmap = sns.diverging_palette(230, 20, as_cmap=True) # Draw the heatmap with the mask and correct aspect ratio sns.set_style(rc = {'figure.facecolor': '#1C1D1F'}) sns.heatmap(corr, cmap=cmap, vmax=.3, center=0, square=True, linewidths=.5, cbar_kws={"shrink": .5}, annot = True)