filtered_df = df[(df['age'] > 30) & (df['city'] == 'New York')]

#44. How can you find the number of unique values in a column?
A: Use the .nunique() method.

unique_cities_count = df['city'].nunique()

#45. What is the difference between a pandas Series and a DataFrame?
A:
• A Series is a one-dimensional labeled array, capable of holding any data type. It's like a single column in a spreadsheet.
• A DataFrame is a two-dimensional labeled data structure with columns of potentially different types. It's like a whole spreadsheet or an SQL table.

#46. How do you sort a DataFrame by the 'salary' column in descending order?
A: Use .sort_values().

sorted_df = df.sort_values(by='salary', ascending=False)

#47. What is method chaining in pandas?
A: Method chaining is the practice of calling methods on a DataFrame sequentially. It improves code readability by reducing the need for intermediate variables.

# Example of method chaining
result = (df[df['age'] > 30]
          .groupby('department')
          ['salary']
          .mean()
          .sort_values(ascending=False))

#48. How do you rename the column 'user_name' to 'username'?
A: Use the .rename() method.

df.rename(columns={'user_name': 'username'}, inplace=True)

#49. How do you get the correlation matrix for all numerical columns in a DataFrame?
A: Use the .corr() method.

correlation_matrix = df.corr(numeric_only=True)

#50. When would you use NumPy over pandas?
A:
• Use NumPy for performing complex mathematical operations on numerical data, especially in machine learning, where data is often represented as arrays (matrices). It is faster for numerical computations.
• Use pandas when you need to work with tabular data, handle missing values, use labeled axes, and perform data manipulation, cleaning, and preparation tasks. Pandas is built on top of NumPy.

---
Part 3: Statistics & Probability Questions (Q51-65)

#51. What is the difference between mean, median, and mode?
A:
• Mean: The average of all data points. It is sensitive to outliers.
• Median: The middle value of a dataset when sorted. It is robust to outliers.
• Mode: The most frequently occurring value in a dataset.

#52. Explain p-value.
A: The p-value is the probability of observing results as extreme as, or more extreme than, what was actually observed, assuming the null hypothesis is true. A small p-value (typically ≤ 0.05) indicates strong evidence against the null hypothesis, so you reject it.

#53. What are Type I and Type II errors?
A:
• Type I Error (False Positive): Rejecting the null hypothesis when it is actually true. (e.g., concluding a new drug is effective when it is not).
• Type II Error (False Negative): Failing to reject the null hypothesis when it is actually false. (e.g., concluding a new drug is not effective when it actually is).

#54. What is a confidence interval?
A: A confidence interval is a range of values, derived from sample statistics, that is likely to contain the value of an unknown population parameter. For example, a 95% confidence interval means that if we were to repeat the experiment many times, 95% of the calculated intervals would contain the true population parameter.

#55. Explain the Central Limit Theorem (CLT).
A: The Central Limit Theorem states that the sampling distribution of the sample mean will be approximately normally distributed, regardless of the original population's distribution, as long as the sample size is sufficiently large (usually n > 30). This is fundamental to hypothesis testing.

#56. What is the difference between correlation and causation?
A:
• Correlation is a statistical measure that indicates the extent to which two or more variables fluctuate together. A positive correlation indicates that the variables increase or decrease in parallel; a negative correlation indicates that as one variable increases, the other decreases.
• Causation indicates that one event is the result of the occurrence of the other event; i.e., there is a causal relationship between the two events. Correlation does not imply causation.

#57. What is A/B testing?
A: A/B testing is a randomized experiment with two variants, A and B. It is a method of comparing two versions of a webpage, app, or feature against each other to determine which one performs better. A key metric is chosen (e.g., click-through rate), and statistical tests are used to determine if the difference in performance is statistically significant.

#58. What are confounding variables?
A: A confounding variable is an "extra" variable that you didn't account for. It can ruin an experiment and give you useless results because it is related to both the independent and dependent variables, creating a spurious association.

#59. What is selection bias?
A: Selection bias is the bias introduced by the selection of individuals, groups, or data for analysis in such a way that proper randomization is not achieved, thereby ensuring that the sample obtained is not representative of the population intended to be analyzed.

#60. You are rolling two fair six-sided dice. What is the probability of rolling a sum of 7?
A:
• Total possible outcomes: 6 * 6 = 36.
• Favorable outcomes for a sum of 7: (1,6), (2,5), (3,4), (4,3), (5,2), (6,1). There are 6 favorable outcomes.
• Probability = Favorable Outcomes / Total Outcomes = 6 / 36 = 1/6.

---
#61. What is standard deviation and variance?
A:
• Variance measures how far a set of numbers is spread out from their average value. It is the average of the squared differences from the mean.
• Standard Deviation is the square root of the variance. It is expressed in the same units as the data, making it more interpretable than variance. It quantifies the amount of variation or dispersion of a set of data values.

#62. Explain conditional probability.
A: Conditional probability is the probability of an event occurring, given that another event has already occurred. It is denoted as P(A|B), the probability of event A given event B. The formula is P(A|B) = P(A and B) / P(B).

#63. What is the law of large numbers?
A: The law of large numbers is a theorem that states that as the number of trials of a random process increases, the average of the results obtained from a large number of trials should be close to the expected value, and will tend to become closer as more trials are performed.

#64. What is a normal distribution? What are its key properties?
A: A normal distribution, also known as a Gaussian distribution, is a probability distribution that is symmetric about the mean, showing that data near the mean are more frequent in occurrence than data far from the mean.
• Properties: It is bell-shaped, symmetric, and defined by its mean (μ) and standard deviation (σ). About 68% of data falls within one standard deviation of the mean, 95% within two, and 99.7% within three.

#65. What is regression analysis? What are some types?
A: Regression analysis is a set of statistical processes for estimating the relationships between a dependent variable (often called the 'outcome variable') and one or more independent variables (often called 'predictors' or 'features').
• Types: Linear Regression, Logistic Regression, Polynomial Regression, Ridge Regression.

---
Part 4: Product Sense & Case Study Questions (Q66-80)

#66. How would you measure the success of a new feature on Instagram, like 'Reels'?
A: My approach would be:
• Define Goals: What is the feature for? Increased user engagement, attracting new users, competing with TikTok.
• Identify Key Metrics (HEART framework):
- Happiness: User surveys, App Store ratings.
- Engagement: Daily Active Users (DAU) creating/viewing Reels, average view time per user, shares, likes, comments.
- Adoption: Number of new users trying the feature, percentage of DAU using Reels.
- Retention: Does using Reels make users more likely to return to the app? (Cohort analysis).
- Task Success: How easy is it to create a Reel? (Time to create, usage of editing tools).
• Counter Metrics: We must also check for negative impacts, like a decrease in time spent on the main feed or Stories.

#67. The number of daily active users (DAU) for our product dropped by 10% yesterday. How would you investigate?
A: I'd follow a structured process:
• Clarify & Validate: Is the data correct? Is it a tracking error or a real drop? Check dashboards and data pipelines. Is the drop global or regional? Affecting all platforms (iOS, Android, Web)?
• Internal Factors (Our Fault):
- Did we just release a new version? Check for bugs.
- Did we have a server outage? Check system health dashboards.
- Did a marketing campaign just end?
• External Factors (Not Our Fault):
- Was it a major holiday?
- Was there a major news event or competitor launch?
- Is it part of a weekly/seasonal trend? (Compare to last week/year).
• Segment the Data: If it's a real drop, I'd segment by user demographics (new vs. returning users, country, device type) to isolate the source of the drop.

#68. How would you design an A/B test for changing a button color from blue to green on our website's homepage?
A:
• Hypothesis: Changing the button color from blue to green will increase the click-through rate (CTR) because green is often associated with "go" and stands out more.
• Control (A): The current blue button.
• Variant (B): The new green button.
• Key Metric: The primary metric is Click-Through Rate (CTR) = (Number of clicks / Number of impressions).
• Setup: Randomly assign users into two groups. 50% see the blue button (control), 50% see the green button (variant).
• Duration & Sample Size: Calculate the required sample size based on the current CTR and the desired minimum detectable effect. Run the test long enough (e.g., 1-2 weeks) to account for weekly variations.
• Conclusion: After the test, use a statistical test (like a Chi-Squared test) to check if the difference in CTR is statistically significant (p-value < 0.05).

#69. What metrics would you use to evaluate a subnoscription-based product like Netflix?
A:
• Acquisition: Customer Acquisition Cost (CAC), Sign-ups.
• Engagement: Daily/Monthly Active Users (DAU/MAU), Average viewing hours per user, Content diversity (number of different shows watched).
• Retention: Churn Rate (monthly/annually), Customer Lifetime Value (CLV).
• Monetization: Monthly Recurring Revenue (MRR), Average Revenue Per User (ARPU).

#70. Our company wants to launch a food delivery service in a new city. What data would you analyze to decide if we should?
A:
• Market Size & Demographics: Population density, age distribution, average income. Is there a large enough target audience?
• Competition: Who are the existing competitors (Uber Eats, DoorDash)? What is their market share, pricing, and restaurant coverage?
• Restaurant Data: Number and type of restaurants in the area. Are there enough high-demand restaurants willing to partner?
• Logistics: Analyze traffic patterns and geographical layout to estimate delivery times and costs.
• Surveys: Conduct surveys to gauge consumer interest and price sensitivity.
---
#71. How do you decide whether a change in a metric is due to seasonality or a real underlying trend?
A:
• Time Series Analysis: Decompose the time series into trend, seasonality, and residual components.
• Year-over-Year Comparison: Compare the metric to the same period in the previous year (e.g., this Monday vs. last year's same Monday). This helps control for seasonality.
• Moving Averages: Use moving averages to smooth out short-term fluctuations and highlight longer-term trends.
• Statistical Tests: Use statistical tests to see if the change is significant after accounting for seasonal effects.

#72. What are the key differences between a data warehouse and a data lake?
A:
• Data Structure: A data warehouse stores structured, processed data. A data lake stores raw data in its native format (structured, semi-structured, unstructured).
• Purpose: Warehouses are designed for business intelligence and reporting. Lakes are used for data exploration, and machine learning.
• Schema: Warehouses use a "schema-on-write" approach (data is structured before being loaded). Lakes use "schema-on-read" (structure is applied when the data is pulled for analysis).

#73. Explain ETL (Extract, Transform, Load).
A: ETL is a data integration process.
• Extract: Data is extracted from various source systems (databases, APIs, logs).
• Transform: The extracted data is cleaned, validated, and transformed into the proper format or structure for the target system (e.g., converting data types, aggregating data).
• Load: The transformed data is loaded into the destination, which is often a data warehouse.

#74. How would you explain a technical concept like standard deviation to a non-technical stakeholder?
A: "Standard deviation is a simple way to measure how spread out our data is. Imagine we're looking at customer ages. A low standard deviation means most of our customers are around the same age, clustered close to the average. A high standard deviation means our customers' ages are very spread out, ranging from very young to very old. It helps us understand the consistency of a dataset."

#75. What is the value of data visualization?
A:
• Clarity: It simplifies complex data, making it easier to understand patterns, trends, and outliers.
• Storytelling: It allows you to tell a compelling story with your data, making your findings more impactful.
• Efficiency: Humans can process visual information much faster than tables of numbers. It helps in quickly identifying relationships and insights.
• Accessibility: It makes data accessible and understandable to a wider, non-technical audience.

---
#76. Our user engagement metric is flat. What could this mean and how would you investigate?
A: Flat engagement can be both good and bad.
• Good: It could mean we have a mature, stable product with a loyal user base.
• Bad: It could mean we are failing to attract new users or that existing users are losing interest (stagnation).
• Investigation:
1. Segment: Break down the overall metric. Is engagement flat for all user segments (new vs. old, by country, by platform)? You might find that new user engagement is rising while old user engagement is falling, resulting in a flat overall number.
2. Feature Usage: Analyze engagement with specific features. Are users shifting their behavior from one feature to another?
3. Competitive Analysis: Look at what competitors are doing. Is the entire market flat?

#77. What is a "good" data visualization? What are some common mistakes?
A:
• Good Visualization: It is accurate, easy to understand, tells a clear story, and is not misleading. It has clear labels, a noscript, and an appropriate chart type for the data (e.g., line chart for time series, bar chart for comparison).
• Common Mistakes:
- Using the wrong chart type.
- Misleading axes (e.g., not starting a bar chart at zero).
- "Chart junk": too many colors, 3D effects, or unnecessary elements that distract from the data.
- Lack of context or clear labels.

#78. What's more important: having perfect data or getting quick insights from imperfect data?
A: It depends on the context.
• For financial reporting or critical system decisions, data accuracy is paramount.
• For initial exploratory analysis, identifying trends, or making quick business decisions, getting timely insights from reasonably good data is often more valuable. The goal is often directionally correct insights to inform the next step, not perfect precision. It's a trade-off between speed and accuracy.

#79. How do you handle outliers in a dataset?
A:
• Identify: Use visualization (box plots, scatter plots) or statistical methods (Z-score, IQR method) to detect them.
• Investigate: Understand why they exist. Are they data entry errors, or are they legitimate but extreme values?
• Handle:
- Remove: If they are errors, they can be removed.
- Transform: Apply a transformation (like log transformation) to reduce their impact.
- Cap: Cap the values at a certain threshold (e.g., replace all values above the 99th percentile with the 99th percentile value).
- Keep: If they are legitimate and important (e.g., fraudulent transactions), they should be kept and studied separately.

#80. Our CEO wants to know the "average session duration" for our app. What are the potential pitfalls of this metric?
A:
• It's an average: It can be heavily skewed by a small number of users with extremely long sessions (outliers). The median session duration might be a more robust metric.
• Doesn't measure quality: A long session isn't necessarily a good session. A user could be struggling to find something, or they could have left the app open in the background.
• Definition is key: How do we define the end of a session? 30 minutes of inactivity? This definition can significantly change the metric.
• I would present the median alongside the mean and also provide a distribution of session lengths to give a more complete picture.

---
Part 5: Technical & Behavioral Questions (Q81-100)

#81. What tools are you proficient in for data analysis?
A: I am highly proficient in SQL for data extraction and querying from relational databases like PostgreSQL and MySQL. For data cleaning, manipulation, analysis, and visualization, I primarily use Python with libraries such as pandas, NumPy, Matplotlib, and Seaborn. I also have experience using BI tools like Tableau/Power BI for creating interactive dashboards.

#82. Describe a data analysis project you are proud of.
A: Use the STAR method:
• Situation: "At my previous company, we were experiencing a higher-than-expected customer churn rate."
• Task: "I was tasked with identifying the key drivers of churn to help the product team develop a retention strategy."
• Action: "I extracted user activity data, subnoscription information, and customer support tickets using SQL. In Python, I cleaned the data and engineered features like 'days since last login' and 'number of support tickets'. I then performed an exploratory data analysis, built a logistic regression model to identify the most significant predictors of churn, and visualized the findings in a Tableau dashboard."
• Result: "My analysis revealed that customers who didn't use a key feature within their first week were 50% more likely to churn. This insight led the product team to redesign the onboarding flow to highlight this feature, which contributed to a 15% reduction in first-month churn over the next quarter."

#83. How do you ensure the quality of your data and analysis?
A:
• Data Validation: I always start by checking for missing values, duplicates, and outliers. I write validation noscripts and perform exploratory data analysis to understand the data's distribution and sanity-check its values.
• Code Reviews: I ask peers to review my SQL queries and Python noscripts for logic errors and efficiency.
• Documentation: I thoroughly document my methodology, assumptions, and steps so my work is reproducible and transparent.
• Triangulation: I try to verify my findings using different data sources or analytical approaches if possible.

#84. How do you communicate your findings to a non-technical audience?
A: I focus on the "so what" rather than the "how". I start with the key insight or recommendation upfront. I use clear, simple language and avoid jargon. I rely heavily on effective data visualizations to tell the story. For example, instead of saying "the p-value was 0.01," I would say, "the data shows with high confidence that our new feature is increasing user engagement."

#85. Describe a time you made a mistake in your analysis. What did you do?
A: "In one project, I was analyzing marketing campaign performance and initially concluded that a specific campaign had a very high ROI. However, a colleague reviewing my query pointed out that I had forgotten to exclude test user accounts from my analysis. I immediately acknowledged the mistake, re-ran the query with the correct filters, and presented the updated, more modest results to my manager. I learned the importance of having a peer review process for critical queries and now I always build a data validation checklist for every project."
---
#86. What is bias-variance tradeoff?
A: It's a central concept in machine learning.
• Bias is the error from erroneous assumptions in the learning algorithm. High bias can cause a model to miss relevant relations between features and target outputs (underfitting).
• Variance is the error from sensitivity to small fluctuations in the training set. High variance can cause a model to model the random noise in the training data (overfitting).
The tradeoff is that models with low bias tend to have high variance, and vice-versa. The goal is to find a balance that minimizes total error.

#87. What is feature engineering?
A: Feature engineering is the process of using domain knowledge to create new features (predictor variables) from raw data. The goal is to improve the performance of machine learning models. Examples include combining features, creating dummy variables from categorical data, or extracting components from a date.

#88. How would you handle a very large dataset that doesn't fit into your computer's memory?
A:
• Chunking: Read and process the data in smaller chunks using options in pd.read_csv (chunksize).
• Data Types: Optimize data types to use less memory (e.g., using int32 instead of int64).
• Cloud Computing: Use cloud-based platforms like AWS, GCP, or Azure that provide scalable computing resources (e.g., using Spark with Databricks).
• Sampling: Work with a representative random sample of the data for initial exploration.

#89. Where do you go to stay up-to-date with the latest trends in data analysis?
A: I actively read blogs like Towards Data Science on Medium, follow key data scientists and analysts on LinkedIn and Twitter, and listen to podcasts. I also enjoy browsing Kaggle competitions to see how others approach complex problems and occasionally review documentation for new features in libraries like pandas and Scikit-learn.

#90. What is a key performance indicator (KPI)?
A: A KPI is a measurable value that demonstrates how effectively a company is achieving key business objectives. Organizations use KPIs to evaluate their success at reaching targets. For a data analyst, it's crucial to understand what the business's KPIs are in order to align analysis with business goals.
---
#91. What is the difference between structured and unstructured data?
A:
• Structured Data: Highly organized and formatted in a way that is easily searchable in relational databases (e.g., spreadsheets, SQL databases).
• Unstructured Data: Data that has no predefined format or organization, making it more difficult to collect, process, and analyze (e.g., text in emails, images, videos, social media posts).

#92. Why is data cleaning important?
A: Data cleaning (or data cleansing) is crucial because "garbage in, garbage out." Raw data is often messy, containing errors, inconsistencies, and missing values. If this data is not cleaned, it will lead to inaccurate analysis, flawed models, and unreliable conclusions, which can result in poor business decisions.

#93. Tell me about a time you had to work with ambiguous instructions or unclear data.
A: "I was once asked to analyze 'user engagement'. This term was very broad. I scheduled a meeting with the stakeholders (product manager, marketing lead) to clarify. I asked questions like: 'What business question are we trying to answer with this analysis?', 'Which users are we most interested in?', and 'What actions on the platform do we consider valuable engagement?'. This helped us collaboratively define engagement with specific metrics (e.g., likes, comments, session duration), which ensured my analysis was relevant and actionable."

#94. What is the difference between a dashboard and a report?
A:
• Report: A static presentation of data for a specific time period (e.g., a quarterly sales report). It's meant to inform.
• Dashboard: A dynamic, interactive BI tool that provides a real-time, at-a-glance view of key performance indicators. It's meant for monitoring and exploration.

#95. What is statistical power?
A: Statistical power is the probability that a hypothesis test will correctly reject the null hypothesis when the null hypothesis is false (i.e., the probability of avoiding a Type II error). In A/B testing, higher power means you are more likely to detect a real effect if one exists.

#96. How do you know if your sample is representative of the population?
A: The best way is through proper sampling techniques. Random sampling is the gold standard, where every member of the population has an equal chance of being selected. You can also use stratified sampling, where you divide the population into subgroups (strata) and then take a random sample from each subgroup to ensure all groups are represented proportionally.

#97. What is your favorite data visualization and why?
A: "I find the box plot to be incredibly powerful and efficient. In a single, compact chart, it visualizes the distribution of data, showing the median, quartiles (25th and 75th percentiles), and potential outliers. It's excellent for comparing distributions across multiple categories and is much more informative than a simple bar chart of means."

#98. What is survivorship bias?
A: Survivorship bias is a logical error where you concentrate on the people or things that "survived" some process and inadvertently overlook those that did not because of their lack of visibility. A classic example is analyzing the habits of successful startup founders without considering the thousands who failed, which can lead to flawed conclusions about what it takes to succeed.

#99. You are given two datasets. How would you figure out if they can be joined?
A: I would first inspect the columns in both datasets to look for a common key or field. This field should ideally be a unique identifier (like user_id, product_id). I would check that the data types of these key columns are the same. Then, I would check the overlap of values between the key columns to understand how many records would match in a join.

#100. Why do you want to be a data analyst?
A: "I am passionate about being a data analyst because I enjoy the process of transforming raw data into actionable insights that can drive real business decisions. I love the blend of technical skills like SQL and Python with the problem-solving and storytelling aspects of the role. I find it incredibly rewarding to uncover hidden patterns and help a company grow by making data-informed choices."

━━━━━━━━━━━━━━━
By: @DataScienceM ✨

📌 How to Evaluate Search Relevance and Ranking

🗂 Category:

🕒 Date: 2024-05-30 | ⏱️ Read time: 6 min read

This article explores the key metrics used for evaluating Search Relevance and Ranking, empowering you…

📌 Implementing Generative and Analytical Models to Create and Enrich Knowledge Graphs for RAGs

🗂 Category:

🕒 Date: 2024-05-29 | ⏱️ Read time: 20 min read

Evaluate generative and analytical models to build Knowledge Graphs and facilitate them to power highly…

📌 From Classical Models to AI: Forecasting Humidity for Energy and Water Efficiency in Data Centers

🗂 Category: MACHINE LEARNING

🕒 Date: 2025-11-02 | ⏱️ Read time: 27 min read

From ARIMA to N-BEATS: Comparing forecasting approaches that balance accuracy, interpretability, and sustainability

📌 MobileNetV3 Paper Walkthrough: The Tiny Giant Getting Even Smarter

🗂 Category: ARTIFICIAL INTELLIGENCE

🕒 Date: 2025-11-02 | ⏱️ Read time: 29 min read

MobileNetV3 with PyTorch  —  now featuring SE blocks and hard activation functions

💡 Applying Image Filters with Pillow

Pillow's ImageFilter module provides a set of pre-defined filters you can apply to your images with a single line of code. This example demonstrates how to apply a Gaussian blur effect, which is useful for softening images or creating depth-of-field effects.

from PIL import Image, ImageFilter

try:
    # Open an existing image
    with Image.open("your_image.jpg") as img:
        # Apply the Gaussian Blur filter
        # The radius parameter controls the blur intensity
        blurred_img = img.filter(ImageFilter.GaussianBlur(radius=5))

        # Display the blurred image
        blurred_img.show()

        # Save the new image
        blurred_img.save("blurred_image.png")

except FileNotFoundError:
    print("Error: 'your_image.jpg' not found. Please provide an image.")

Code explanation: The noscript opens an image file, applies a GaussianBlur filter from the ImageFilter module using the .filter() method, and then displays and saves the resulting blurred image. The blur intensity is controlled by the radius argument.

#Python #Pillow #ImageProcessing #ImageFilter #PIL

━━━━━━━━━━━━━━━
By: @DataScienceM ✨

💡 Top 50 Operations for Audio Processing in Python

Note: Most examples use pydub. You need ffmpeg installed for opening/exporting non-WAV files. Install libraries with pip install pydub librosa sounddevice scipy numpy.

I. Basic Loading, Saving & Properties

• Load an audio file (any format).

from pydub import AudioSegment
audio = AudioSegment.from_file("sound.mp3")

• Export (save) an audio file.

audio.export("new_sound.wav", format="wav")

• Get duration in milliseconds.

duration_ms = len(audio)

• Get frame rate (sample rate).

rate = audio.frame_rate

• Get number of channels (1 for mono, 2 for stereo).

channels = audio.channels

• Get sample width in bytes (e.g., 2 for 16-bit).

width = audio.sample_width

II. Playback & Recording

• Play an audio segment.

from pydub.playback import play
play(audio)

• Record audio from a microphone for 5 seconds.

import sounddevice as sd
from scipy.io.wavfile import write

fs = 44100  # Sample rate
seconds = 5
recording = sd.rec(int(seconds * fs), samplerate=fs, channels=2)
sd.wait()  # Wait until recording is finished
write('output.wav', fs, recording)

III. Slicing & Concatenating

• Get a slice (e.g., the first 5 seconds).

first_five_seconds = audio[:5000] # Time is in milliseconds

• Get a slice from the end (e.g., the last 3 seconds).

last_three_seconds = audio[-3000:]

• Concatenate (append) two audio files.

combined = audio1 + audio2

• Repeat an audio segment.

repeated = audio * 3

• Crossfade two audio segments.

# Fades out audio1 while fading in audio2
faded = audio1.append(audio2, crossfade=1000)

IV. Volume & Effects

• Increase volume by 6 dB.

louder_audio = audio + 6

• Decrease volume by 3 dB.

quieter_audio = audio - 3

• Fade in from silence.

faded_in = audio.fade_in(2000) # 2-second fade-in

• Fade out to silence.

faded_out = audio.fade_out(3000) # 3-second fade-out

• Reverse the audio.

reversed_audio = audio.reverse()

• Normalize audio to a maximum amplitude.

from pydub.effects import normalize
normalized_audio = normalize(audio)

• Overlay (mix) two tracks.

# Starts playing 'overlay_sound' 5 seconds into 'main_sound'
mixed = main_sound.overlay(overlay_sound, position=5000)

V. Channel Manipulation

• Split stereo into two mono channels.

left_channel, right_channel = audio.split_to_mono()

• Create a stereo segment from two mono segments.

stereo_sound = AudioSegment.from_mono_segments(left_channel, right_channel)

• Convert stereo to mono.

mono_audio = audio.set_channels(1)

VI. Silence & Splitting

• Generate a silent segment.

one_second_silence = AudioSegment.silent(duration=1000)

• Split audio based on silence.

from pydub.silence import split_on_silence
chunks = split_on_silence(
    audio,
    min_silence_len=500,
    silence_thresh=-40
)

VII. Working with Raw Data (NumPy & SciPy)

• Get raw audio data as a NumPy array.

import numpy as np
samples = np.array(audio.get_array_of_samples())

• Create a Pydub segment from a NumPy array.

new_audio = AudioSegment(
    samples.tobytes(),
    frame_rate=audio.frame_rate,
    sample_width=audio.sample_width,
    channels=audio.channels
)

• Read a WAV file directly into a NumPy array.

from scipy.io.wavfile import read
rate, data = read("sound.wav")

• Write a NumPy array to a WAV file.

from scipy.io.wavfile import write
write("new_sound.wav", rate, data)

• Generate a sine wave.

import numpy as np
sample_rate = 44100
frequency = 440 # A4 note
duration = 5
t = np.linspace(0., duration, int(sample_rate * duration))
amplitude = np.iinfo(np.int16).max * 0.5
data = amplitude * np.sin(2. * np.pi * frequency * t)
# This array can now be written to a file

VIII. Audio Analysis with Librosa

• Load audio with Librosa.

import librosa
y, sr = librosa.load("sound.mp3")

• Estimate tempo (Beats Per Minute).

tempo, beat_frames = librosa.beat.beat_track(y=y, sr=sr)

• Get beat event times in seconds.

beat_times = librosa.frames_to_time(beat_frames, sr=sr)

• Decompose into harmonic and percussive components.

y_harmonic, y_percussive = librosa.effects.hpss(y)

• Compute a spectrogram.

import numpy as np
D = librosa.stft(y)
S_db = librosa.amplitude_to_db(np.abs(D), ref=np.max)

• Compute Mel-Frequency Cepstral Coefficients (MFCCs).

mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13)

• Compute Chroma features (related to musical pitch).

chroma = librosa.feature.chroma_stft(y=y, sr=sr)

• Detect onset events (the start of notes).

onset_frames = librosa.onset.onset_detect(y=y, sr=sr)
onset_times = librosa.frames_to_time(onset_frames, sr=sr)

• Pitch shifting.

y_pitched = librosa.effects.pitch_shift(y, sr=sr, n_steps=4) # Shift up 4 semitones

• Time stretching (change speed without changing pitch).

y_fast = librosa.effects.time_stretch(y, rate=2.0) # Double speed

IX. More Utilities

• Detect leading silence.

from pydub.silence import detect_leading_silence
trim_ms = detect_leading_silence(audio)
trimmed_audio = audio[trim_ms:]

• Get the root mean square (RMS) energy.

rms = audio.rms

• Get the maximum possible RMS for the audio format.

max_possible_rms = audio.max_possible_amplitude

• Find the loudest section of an audio file.

from pydub.scipy_effects import normalize
loudest_part = normalize(audio.strip_silence(silence_len=1000, silence_thresh=-32))

• Change the frame rate (resample).

resampled = audio.set_frame_rate(16000)

• Create a simple band-pass filter.

from pydub.scipy_effects import band_pass_filter
filtered = band_pass_filter(audio, 400, 2000) # Pass between 400Hz and 2000Hz

• Convert file format in one line.

AudioSegment.from_file("music.ogg").export("music.mp3", format="mp3")

• Get the raw bytes of the audio data.

raw_data = audio.raw_data

• Get the maximum amplitude.

max_amp = audio.max

• Match the volume of two segments.

matched_audio2 = audio2.apply_gain(audio1.dBFS - audio2.dBFS)

#Python #AudioProcessing #Pydub #Librosa #SignalProcessing

━━━━━━━━━━━━━━━
By: @DataScienceM ✨

📌 The Pearson Correlation Coefficient, Explained Simply

🗂 Category: STATISTICS

🕒 Date: 2025-11-01 | ⏱️ Read time: 7 min read

A simple explanation of the Pearson correlation coefficient with examples

📌 Graph RAG vs SQL RAG

🗂 Category: LARGE LANGUAGE MODELS

🕒 Date: 2025-11-01 | ⏱️ Read time: 7 min read

Evaluating RAGs on graph and SQL databases

📌 Understanding the Two Faces of Shiny for Python: Core and Express

🗂 Category: DATA SCIENCE

🕒 Date: 2024-05-29 | ⏱️ Read time: 7 min read

Exploring the Differences and Use Cases of Shiny Core and Shiny Express for Python

📌 Do You Need a Degree to Be a Data Scientist?

🗂 Category: DATA SCIENCE

🕒 Date: 2024-05-29 | ⏱️ Read time: 8 min read

No, but it certainly helps.