Transcript Karaoke Formation using Matlab

KARAOKE FORMATION
Pratik Bhanawat (10bec113)
Gunjan Gupta (10bec112)
INTRODUCTION

Karaoke is a Japanese abbreviated compound
word, "kara" comes from "karappo" meaning
empty, and "oke" is the abbreviation of
"okesutura," or orchestra. Usually, a recorded
popular song consists of vocals and
accompaniment. Musical works in which only
the accompaniment is recorded were named
"karaoke." Karaoke singing involves singing to
such recorded accompaniments of popular songs
in front of a live audience. After the singer
chooses a song from a catalogue, lyrics are
usually displayed on a monitor, recorded music
plays, and it's Showtime for the novice pop star.
INTRODUCTION (cont.)

Invented in the late 1970's, the wild
popularity of karaoke over the years has
swept this form of singing into the
mainstream throughout the world.
Karaoke creates its own culture, while
reflecting much about the wider culture
and the place of popular music as a media
form.
Brief Of Algorithm Used

Aim - This program attempts to create
karaoke of a song. It attempts to remove
voice and then writes it to a new wav file.

Principle - This program utilizes the fact
that voice is recorded equally in both
channels without any stereo effect.
Brief Of Algorithm Used

Method - First one of the channel is high
pass filtered. This is done to protect the
low end bass which is also recorded
equally on both channels. Then this
channel is subtracted from the other. The
common part that is voice gets cancelled.
Brief Of Algorithm Used

Select the high pass filter cut off about
50-75 Hz. If there is too much voice in the
output reduce this value. If it has too little
bass increase this value. Note that by
doing so voice will also increase.
Drawbacks
Output is mono and a bit harsh .
 Does not remove chorus .
 Also fails if the voice is not identical in
both channels .
 It takes only .wav as input.You can
convert your .mp3 into .wav using any
converter.

So, Let’s Begin
Few Other Algorithms

ALGORITHM – 1
It is a two step process comprising of a training phase,
during which a statistical model is created for a
singer’s voice, and a working phase, during which the
starting point of the singing voice is detected and a
fixed length of testing data is taken from that point.
Audio features extracted from this data are then
compared against the existing singers’ models to
perform singer identification. Singing voice detection
is achieved by extracting features of energy, average
zero-crossing rate (ZCR), harmonic coefficients and
spectral flux computed at regular intervals which are
then compared against a set of predetermined
thresholds.
Few Other Algorithms

ALGORITHM – II
A statistical model to classify segments of
musical audio into vocal or non-vocal using
a Hidden Markov Model (HMM) classifier.
The feature extraction is based on sub-band
processing that uses the log frequency power
coefficients (LFPC) to provide an indication
of the energy distribution among sub-bands.
The training model also takes into account
tempo and song structure information in song
modeling based on the observed variations in
intra-song signal characteristics.
Future Advancement

To Transcribe the lyrics - but this is
currently impractical. Transcription of lyrics
using speech recognition is an extremely
challenging task as singing differs from
speech in many ways. The phonetic and
timing
modification,
presence
of
meaningless syllables often employed by
singers and interference of the instrumental
background would make an acoustic
classifier trained on normal speech a poor
match to the acoustics of the sung vocal line.