Scribd
Upload
4 views

MTP 1

The document provides an overview of sound, including its components such as amplitude, frequency, and tempo, as well as techniques for audio processing like noise reduction and reverb addition. It discusses methods for detecting musical onsets and integrating instruments into audio tracks, emphasizing the importance of rhythm and clarity in music. Future work includes developing online platforms for audio processing and real-time systems for live performances.

Uploaded by

reheyi2494
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
4 views

MTP 1

The document provides an overview of sound, including its components such as amplitude, frequency, and tempo, as well as techniques for audio processing like noise reduction and reverb addition. It discusses methods for detecting musical onsets and integrating instruments into audio tracks, emphasizing the importance of rhythm and clarity in music. Future work includes developing online platforms for audio processing and real-time systems for live performances.

Uploaded by

reheyi2494
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 32

What is Sound?

• It is a combination of various waveform signals aka sinusoids of


different time intervals, frequencies, tempo and amplitude.
• ‘Amplitude’ represent ‘volume’ and frequency represent pitch of the
sound.
• ‘Tempo’ or speed of sound can be defined as rate of change in
frequency.
• Frequency is the number of cycles of a wave or a periodic event that
occur within a specific unit of time.
• If a sound wave has a frequency of 1000 Hz, it means there are 1000
wave cycles every second.
Sound with constant frequency (beep sound) has no
meaning for tempo.

Constant Frequency, amplitude and tempo.

Sinusoidally varying frequency; Constant amplitude and


tempo.
Experiments:
Sinusoidally varying frequency and tempo; Constant
amplitude.

Sinusoidally varying frequency and tempo; Linearly


increasing amplitude

Bike Revving
Key Components of
Sound
• Attack: The initial burst of sound when a note
starts, often sharp and loud.
• Transient: The short, sudden increase in sound
that marks the beginning of the note.
• Decay: The gradual fading out of the sound after
the attack.

• References: Juan Pablo Bello, Laurent Daudet, Samer Abdallah, Chris Duxbury, Mike Davies, and Mark
B. Sandler, Senior Member, IEEE
Convert Convert Audio to an Array

Normalize Normalize the Signal Amplitude (-1 to 1)

Onset Segment Segment into Overlapping Frames (30ms-50ms)

Detection
Calculate Calculate Energy for Each Frame (Squared Amplitude)

Detect Detect Sudden Changes in Energy to Identify Peaks


Waveform &
Energy of Audio
Onsets of Audio
Duplicate Duplicate the original audio array to create a new array.

Reduce Reduce the amplitude of the duplicate array (e.g., scale it by 0.5).

Enhancing
Audio with Shift Shift the duplicate array to the right by 0.1 seconds.

Reverb Padding Add padding to the right-shifted and original arrays to ensure they
have the same length.

Combine Combine the two audio arrays into a single merged audio track.
Enhancing Audio with Reverb
Results of Merging Audio File
Noise Reduction

Audio with Noise

Audio after Noise Removal


Extract Extract the noise profile from the audio.

Connvert the audio signal from the time domain to the frequency
Convert domain.

Noise Calculate Calculate the power (magnitude squared) of both the noise and

Reduction
audio across all frequencies.

Subtract the noise power from the audio power at each


Subtract frequency.

Reconstruct the cleaned audio by converting it back from the


Reconstruct frequency domain to the time domain.
Frequency Domain
Extract Extract the noise profile from the audio.

Connvert the audio signal from the time domain to the frequency
Convert domain.

Noise Calculate Calculate the power (magnitude squared) of both the noise and

Reduction
audio across all frequencies.

Subtract the noise power from the audio power at each


Subtract frequency.

Reconstruct the cleaned audio by converting it back from the


Reconstruct frequency domain to the time domain.
Spectrogram
Added Instrument on each Onset.

Added instruments at specific onsets, based on


a 90% amplitude threshold.

Instruments Avoided overlap of instruments by tracking the


end time of instrument.
Sound Integration
Added 4 Instruments in a Loop

Added 4 Instruments A/C to Amplitude


Instrument on Each
Onset

• Detected Onset
• Generated Drum Sound
• Added drum sound to Onset
Added instruments at
specific onsets, based on a
90% amplitude threshold.

• Detected Onset
• Generated Drum Sound
• Added the drum sound to
onsets with the top 90%
amplitude.
Avoided overlap of
instruments

• Detected Onset
• Used Jhanjh instrument
recorded sound
• Tracked the onset start time and
end time of each instrument.
• This ensured clear separation
and precise timing between
different instrumental sounds.
Added 4 Instruments
in a Loop

• Intruments as: Tabla, Ghungroo,


Damru, Jhanjh
• Tracked the onset start time and
end time of each instrument.
• This ensured clear separation
and precise timing between
different instrumental sounds.
Added 4 Instruments
A/C to Amplitude

• Initially, I stored the amplitude of


each onset and sorted the data
in ascending order.
• The data was then divided into
four segments.
• Each segment was assigned to a
different instrument, ensuring
balanced distribution.
Ankur Satyam

Experimented With Friends Audio


Detecting Mukhada in Audio File

Importance:
Mukhada: A repeating Identifying Mukhada
melodic phrase in a helps in rhythmic
song. enhancement and
remixing.
Detecting Mukhada
Manually extracted the Mukhada segment from the song.

Modified Cross-correlation is used to measure the similarity between two signals.

Identify where the Mukhada appears in the song by comparing it to the full
audio.
Iterates over possible shifts, calculating a similarity score.

Experimented with different values of "skip" to optimize the detection.


High
Downsampling
2000

• Computationally
efficient but less
accurate.
• It took 1-2 hours to
compute.
Moderate
Downsampling
500
• Improved accuracy at a
slight increase in
computational cost
compared to the first.
• It took 3-4 hours to
compute.
Low
Downsampling
10
• Computationally
expensive, but most
accurate similarity.
• It took 26-28 hours to
compute.
Conclusion
Onset Detection: Accurate identification of musical onsets helps in rhythm analysis and adding
effects precisely.

Reverb Addition: Adding reverb creates a sense of space and depth, enriching the overall listening
experience.

Noise Reduction: Effective noise reduction techniques improve audio clarity without
compromising quality.

Instrument Integration: Adding extra sounds, like tabla, ghungroo, damru & jhan, helps improve
the rhythm and feel of the music.

Allows for the precise identification of key recurring themes (Mukhda) within a song.
Future Work

Detecting Periodicity in Music: Developing methods to find patterns in a


song and use this to add beats at regular intervals.

Online Audio Processing Platform: Create a website where users can


upload their recorded songs and choose options like noise removal, adding
reverb, or integrating beats and instruments.

Real-Time Model for Live Performances: Develop a real-time system that


can generate instruments beats during live music performances.
References
• Juan Pablo Bello, Laurent Daudet, Samer Abdallah, Chris Duxbury,
Mike Davies, and Mark B. Sandler, Senior Member, IEEE
• D. Sinha, S. Saeed, and A. Ferreira, “A novel automatic noise
removal technique for audio and speech signals,”
• Dattorro jon, “Effect design, part 1: reverberator and other filters,”
journal of the audio engineering society,

You might also like