Codec 2

David Rowe

Revision: d104a50 on branch: pr/38

December 18, 2023

1 Introduction
Codec 2 is an open source speech codec designed for communications quality
speech between 700 and 3200 bit/s. The main application is low bandwidth
HF/VHF digital radio. It fills a gap in open source voice codecs beneath 5000
bit/s and is released under the GNU Lesser General Public License (LGPL).
Key feature includes:

1. A range of modes supporting different bit rates, currently (Nov 2023):

3200, 2400, 1600, 1400, 1300, 1200, 700C. The number is the bit rate, and
the supplementary letter the version (700C replaced the earlier 700, 700A,
700B versions). These are referred to as “Codec 2 3200”, “Codec 2 700C”
etc.

2. Modest CPU (a few 10s of MIPs) and memory (a few 10s of kbytes of
RAM) requirements such that it can run on stm32 class microcontrollers
with hardware FPU.
3. Codec 2 has been designed for digital voice over radio applications, and
retains intelligible speech at a few percent bit error rate.

4. An open source reference implementation in the C language for C99/gcc

compilers, and a cmake build and test framework that runs on Linux.
Also included is a cross compiled stm32 reference implementation.
5. Ports to non-C99 compilers (e.g. MSVC, some microcontrollers, native
builds on Windows) are left to third party developers - we recommend the
tests also be ported and pass before considering the port successful.
The Codec 2 project was started in 2009 in response to the problem of
closed source, patented, proprietary voice codecs in the sub-5 kbit/s range, in
particular for use in the Amateur Radio service.
This document describes Codec 2 at two levels. Section 2 is a high level de-
scription aimed at the Radio Amateur, while Section 3 contains a more detailed

1
description using math and signal processing theory. Combined with the C
source code, it is intended to give the reader enough information to understand
the operation of Codec 2 in detail and embark on source code level projects,
such as improvements, ports to other languages, student or academic research
projects. Issues with the current algorithms and topics for further work are also
included. Section 4 provides a summary of the Codec 2 modes, and Section 5
a guide to the C source files. A glossary of terms and symbols is provided in
Section 6, and Section 7 has suggestions for further documentation work.
This production of this document was kindly supported by an ARDC grant
[1]. As an open source project, many people have contributed to Codec 2 over
the years - we deeply appreciate all of your support.

2 Codec 2 for the Radio Amateur

2.1 Model Based Speech Coding
A speech codec takes speech samples from an A/D converter (e.g. 16 bit samples
at 8 kHz or 128 kbits/s) and compresses them down to a low bit rate that can
be more easily sent over a narrow bandwidth channel (e.g. 700 bits/s for HF).
Speech coding is the art of “what can we throw away”. We need to lower the
bit rate of the speech while retaining speech you can understand, and making
it sound as natural as possible.
As such low bit rates we use a speech production “model”. The input speech
is anlaysed, and we extract model parameters, which are then sent over the
channel. An example of a model based parameter is the pitch of the person
speaking. We estimate the pitch of the speaker, quantise it to a 7 bit number,
and send that over the channel every 20ms.
The model based approach used by Codec 2 allows high compression, with
some trade offs such as noticeable artefacts in the decoded speech. Higher bit
rate codecs (above 5000 bit/s), such as those use for mobile telephony or voice
on the Internet, tend to pay more attention to preserving the speech waveform,
or use a hybrid approach of waveform and model based techniques. They sound
better but require a higher bit rate.
Recently, machine learning has been applied to speech coding. This tech-
nology promises high quality, artefact free speech quality at low bit rates, but
currently (2023) requires significantly more memory and CPU resources than
traditional speech coding technology such as Codec 2. However the field is pro-
gressing rapidly, and as the cost of CPU and memory decreases (Moore’s law)
will soon be a viable technology for many low bit rate speech applications.

2.2 Speech in Time and Frequency

To explain how Codec 2 works, let’s look at some speech. Figure 1 shows a short
40ms segment of speech in the time and frequency domain. On the time plot
we can see the waveform is changing slowly over time as the word is articulated.

2
On the right hand side it also appears to repeat itself - one cycle looks very
similar to the last. This cycle time is the “pitch period”, which for this example
is around P = 35 samples. Given we are sampling at Fs = 8000 Hz, the pitch
period is P/Fs = 35/8000 = 0.0044 seconds, or 4.4ms.

Figure 1: A 40ms segment from the word “these” from a female speaker,
sampled at 8kHz. Top is a plot against time, bottom (blue) is a plot of the same
speech against frequency. The waveform repeats itself every 4.3ms (F0 = 230
Hz); this is the “pitch period” of this segment. The red crosses are the sine
wave amplitudes, explained in the text.

30000

20000

10000
Amplitude

-10000

-20000

-30000
50 100 150 200 250 300
Time (samples)
80

60
Amplitude (dB)

40

20

1000 2000 3000 4000

Frequency (Hz)

3
 Now if the pitch period is 4.4ms, the pitch frequency or fundamental fre-
quency F0 is about 1/0.0044 ≈ 230 Hz. If we look at the blue frequency domain
plot at the bottom of Figure 1, we can see spikes that repeat every 230 Hz.
If the signal is repeating itself in the time domain, it also repeats itself in the
frequency domain. Those spikes separated by about 230 Hz are harmonics of
the fundamental frequency F0 .
Note that each harmonic has its own amplitude, that varies across frequency.
The red line plots the amplitude of each harmonic. In this example, there is
a peak around 500 Hz and another broader peak around 2300 Hz. The ear
perceives speech by the location of these peaks and troughs.

2.3 Sinusoidal Speech Coding

A sinewave will cause a spike or spectral line on a spectrum plot, so we can
see each spike as a small sine wave generator. Each sine wave generator has its
own frequency that are all multiples of the fundamental pitch frequency (e.g.
230, 460, 690, ... Hz). They will also have their own amplitude and phase. If we
add all the sine waves together (Figure 2) we can produce reasonable quality
synthesised speech. This is called sinusoidal speech coding and is the speech
production “model” at the heart of Codec 2.

Figure 2: The sinusoidal speech model. If we sum a series of sine waves, we can
generate a speech signal. Each sinewave has its own amplitude (A1 , A2 , ...AL ),
frequency, and phase (not shown). We assume the frequencies are multiples of
the fundamental frequency F0 . L is the total number of sinewaves we can fit in
4 kHz.

A1 , F0 Hz

A2 , 2F0 Hz

AL , LF0 Hz

4
 The model parameters evolve over time, but can generally be considered
constant for a short time window (a few 10s of ms). For example, pitch evolves
over time, moving up or down as a word is articulated.
As the model parameters change over time, we need to keep updating them.
This is known as the frame rate of the codec, which can be expressed in terms
of frequency (Hz) or time (ms). For sampling model parameters, Codec 2 uses
a frame rate of 10ms. For transmission over the channel, we reduce this to
20-40ms in order to lower the bit rate. The trade off with a lower frame rate is
reduced speech quality.
The parameters of the sinusoidal model are:
1. The frequency of each sine wave. As they are all harmonics of F0 we can
just send F0 to the decoder, and it can reconstruct the frequency of each
harmonic as F0 , 2F0 , 3F0 , ..., LF0 . We used 5-7 bits/frame to represent F0
in Codec 2.
2. The amplitude of each sine wave, A1 , A2 , ..., AL . These “spectral ampli-
tudes” are really important as they convey the information the ear needs
to understand speech. Most of the bits are used for spectral amplitude
information. Codec 2 uses between 18 and 50 bits/frame for spectral am-
plitude information.
3. Voicing information. Speech can be approximated into voiced speech (vow-
els) and unvoiced speech (like consonants), or some mixture of the two.
The example in Figure 1 above is voiced speech. So we need some way to
describe voicing to the decoder. This requires just a few bits/frame.
4. The phase of each sine wave. Codec 2 discards the phases of each harmonic
at the encoder and reconstructs them at the decoder using an algorithm,
so no bits are required for phases. This results in some drop in speech
quality.

2.4 Codec 2 Encoder and Decoder

This section explains how the Codec 2 encoder and decoder work using block
diagrams.
The encoder is presented in Figure 3. Frames of input speech samples are
passed to a Fast Fourier Transform (FFT), which converts the time domain
samples to the frequency domain. The same frame of input samples is used to
estimate the pitch of the current frame. We then use the pitch and frequency
domain speech to estimate the amplitude of each sine wave.
Yet another algorithm is used to determine if the frame is voiced or unvoiced.
This works by comparing the spectrum to what we would expect for voiced
speech (e.g. lots of spectral lines). If the energy is more random and continuous
rather than discrete lines, we consider it unvoiced.
Up until this point the processing happens at a 10ms frame rate. However, in
the next step, we “decimate“ the model parameters - this means we discard some

5
 Figure 3: Codec 2 Encoder.

Input Pitch
Speech Estimator

Estimate Decimate Bit

FFT
Amplitudes Quantise Stream

Estimate
Voicing

of the model parameters to lower the frame rate, which helps us lower the bit
rate. Decimating to 20ms (throwing away every 2nd set of model parameters)
doesn’t have much effect, but beyond that the speech quality starts to degrade.
So there is a trade off between decimation rate and bit rate over the channel.
Once we have the desired frame rate, we “quantise” each model parameter.
This means we use a fixed number of bits to represent it, so we can send the bits
over the channel. Parameters like pitch and voicing are fairly easy, but quite
a bit of DSP goes into quantising the spectral amplitudes. For the higher bit
rate Codec 2 modes, we design a filter that matches the spectral amplitudes,
then send a quantised version of the filter over the channel. Using the example
in Figure 1 - the filter would have a band pass peaks at 500 and 2300 Hz. Its
frequency response would follow the red line. The filter is time varying - we
redesign it for every frame.
You’ll notice the term “estimate” being used a lot. One of the problems
with model based speech coding is the algorithms we use to extract the model
parameters are not perfect. Occasionally the algorithms get it wrong. Look
at the red crosses on the bottom plot of Figure 1. These mark the amplitude
estimate of each harmonic. If you look carefully, you’ll see that above 2000Hz,
the crosses fall a little short of the exact centre of each harmonic. This is an
example of a “fine” pitch estimator error, a little off the correct value.
Often the errors interact, for example the fine pitch error shown above will
mean the amplitude estimates are a little bit off as well. Fortunately, these errors
tend to be temporary and are sometimes not even noticeable to the listener -
remember this codec is often used for HF/VHF radio where channel noise is
part of the normal experience.
Figure 4 shows the operation of the Codec 2 decoder. We take the sequence
of bits received from the channel and recover the quantised model parameters,
pitch, spectral amplitudes, and voicing. We then resample the model parameters

6
 Figure 4: Codec 2 Decoder

Synthesise
Phases

Bit Dequantise Recover Synthesise Output

Stream Interpolate Amplitudes Speech Speech

back up to the 10ms frame rate using a technique called interpolation. For
example, say we receive a F0 = 200 Hz pitch value, then 20ms later F0 = 220
Hz. We can use the average F0 = 210 Hz for the middle 10ms frame.
The phases of each harmonic are generated using the other model parameters
and some DSP. It turns out that if you know the amplitude spectrum, you can
determine a “reasonable” phase spectrum using some DSP operations, which in
practice is implemented with a couple of FFTs. We also use the voicing infor-
mation - for unvoiced speech we use random phases (a good way to synthesise
noise-like signals) - and for voiced speech we make sure the phases are chosen
so the synthesised speech transitions smoothly from one frame to the next.
Frames of speech are synthesised using an inverse FFT. We take a blank array
of FFT samples, and at intervals of F0 insert samples with the amplitude and
phase of each harmonic. We then inverse FFT to create a frame of time domain
samples. These frames of synthesised speech samples are carefully aligned with
the previous frame to ensure smooth frame-frame transitions and output to the
listener.

2.5 Bit Allocation

Table 2.5 presents the bit allocation for two popular Codec 2 modes. One addi-
tional parameter is the frame energy, which is the average level of the spectral
amplitudes, or “AF gain” of the speech frame.
At very low bit rates such as 700 bits/s, we use Vector Quantisation (VQ)
to represent the spectral amplitudes. We construct a table such that each row
of the table has a set of spectral amplitude samples. In Codec 2 700C the table
has 512 rows. During the quantisation process, we choose the table row that
best matches the spectral amplitudes for this frame, then send the index of the
table row. The decoder has a similar table, so can use the index to look up the
spectral amplitude values. If the table is 512 rows, we can use a 9 bit number
to quantise the spectral amplitudes. In Codec 2 700C, we use two tables of 512
entries each (18 bits total), the second one helps fine tune the quantisation from
the first table.
Vector Quantisation can only represent what is present in the tables, so if it

7
sees anything unusual (for example, a different microphone frequency response
or background noise), the quantisation can become very rough and speech qual-
ity poor. We train the tables at design time using a database of speech samples
and a training algorithm - an early form of machine learning.
Codec 2 3200 uses the method of fitting a filter to the spectral amplitudes,
this approach tends to be more forgiving of small variations in the input speech
spectrum, but is not as efficient in terms of bit rate.

Parameter 3200 700C

Pitch F0 7 5
Spectral Amplitudes {Am } 50 18
Energy 5 3
Voicing 2 1
Bits/frame 64 28
Frame Rate 20ms 40ms
Bit rate 3200 700

Table 1: Bit allocation of the 3200 and 700C modes

3 Detailed Design
3.1 Overview
Codec 2 is based on sinusoidal [8] and Multi-Band Excitation (MBE) [3] vocoders
that were first developed in the late 1980s. Descendants of the MBE vocoders
(IMBE, AMBE etc) have enjoyed widespread use in many applications such
as VHF/UHF handheld radios and satellite communications. In the 1990s the
author studied sinusoidal speech coding [10], which provided the skill set and a
practical, patent free baseline for starting the Codec 2 project:
Some features of the Codec 2 Design:
1. A pitch estimator based on a 2nd order non-linearity developed by the
author.
2. A single voiced/unvoiced binary voicing model.
3. A frequency domain IFFT/overlap-add synthesis model for voiced and
unvoiced speech.
4. Phases are not transmitted, they are synthesised at the decoder from the
magnitude spectrum and voicing decision.
5. For the higher bit rate modes (1200 to 3200 bits/s), spectral magnitudes
are represented using LPCs extracted from time domain analysis and
scalar LSP quantisation.
6. For Codec 2 700C, vector quantisation of resampled spectral magnitudes
in the log domain.

8
 7. Minimal interframe prediction in order to minimise error propagation and
maximise robustness to channel errors.
8. A post filter that enhances the speech quality of the baseline codec, espe-
cially for low pitched (male) speakers.

3.2 Sinusoidal Analysis

Both voiced and unvoiced speech is represented using a harmonic sinusoidal
model:
L
X
ŝ(n) = Am cos(ω0 mn + θm ) (1)
m=1

where the parameters Am , θm , m = 1...L represent the magnitude and phases

of each sinusoid, ω0 is the fundamental frequency in radians/sample, and L =
⌊π/ω0 ⌋ is the number of harmonics.
Figure 5 illustrates the processing steps in the sinusoidal analysis system at
the core of the Codec 2 encoder. The algorithms described in this section are
based on the work in [10], with some changes in notation.

Figure 5: Sinusoidal Analysis

Est Amp {Am }

s(n) Window DFT
sw (n) Sω (k) and Phase {θm }

NLP ω0

The time domain speech signal s(n) is divided into overlapping analysis
windows (frames) of Nw = 279 samples. The centre of each analysis window
is separated by N = 80 or 10ms. Codec 2 operates at an internal frame rate
of 100 Hz. To analyse the l-th frame it is convenient to convert the fixed time
reference to a sliding time reference centred on the current analysis window:

sw (n) = s(lN + n)w(n), n = −Nw2 ...Nw2 (2)

where w(n) is a tapered even window of Nw (Nw odd) samples with:

 
Nw
Nw2 = (3)
2
A suitable window function is a shifted Hann window:
 
1 1 2π(n − Nw2 )
w(n) = − cos (4)
2 2 Nw − 1

9
where the energy in the window is normalised such that:
w −1
NX
1
w2 (n) = (5)
n=0
Ndf t

To analyse s(n) in the frequency domain the Ndf t point Discrete Fourier Trans-
form (DFT) can be computed:
N
Xw2

Sw (k) = sw (n)e−j2πkn/Ndf t (6)

n=−Nw2

The magnitude and phase of each harmonic is given by:

v
u bm −1
uX
Am = t |Sw (k)|2 (7)
k=am

θm = arg [Sw (⌊mr⌉] (8)

where:
am = ⌊(m − 0.5)r⌉
bm = ⌊(m + 0.5)r⌉ (9)
ω0 Ndf t
r=
2π
The DFT indexes am , bm select the band of Sw (k) containing the m-th harmonic;
r maps the harmonic number m to the nearest DFT index, and ⌊x⌉ is the
rounding operator. This method of estimating Am is relatively insensitive to
small errors in F 0 estimation and works equally well for voiced and unvoiced
speech. Figure 1 plots Sw (blue) and {Am } (red) for a sample frame of female
speech.
The phase is sampled at the centre of the band. For all practical Codec 2
modes, the phase is not transmitted to the decoder, so it does not need to be
computed. However, speech synthesised using the phase is useful as a control
during development and is available using the c2sim utility.

3.3 Sinusoidal Synthesis

Synthesis is achieved by constructing an estimate of the original speech spectrum
using the sinusoidal model parameters for the current frame. This information
is then transformed to the time domain using an Inverse DFT (IDFT). To
produce a continuous time domain waveform the IDFTs from adjacent frames
are smoothly interpolated using a weighted overlap add procedure [8].
The synthetic speech spectrum is constructed using the sinusoidal model
parameters by populating a DFT array Ŝw (k) with weighted impulses at the

10
Figure 6: Sinusoidal Synthesis. At frame l the windowing function generates
2N samples. The first N samples complete the current frame. The second N
samples are stored for summing with the next frame.

n =0, ..,
ω0 N −1
{Am } Construct Window
IDFT + ŝ(n + lN )
{θm } Sw (k) ŝl (n) t(n)

1 frame
n =N, ..., delay
2N − 1

harmonic centres:
(
Am ejθm , k = ⌊mr⌉, m = 1..L
Ŝw (k) = (10)
0, otherwise

As we wish to synthesise a real time domain signal, Sw (k) is defined to be

conjugate symmetric:
∗
Ŝw (Ndf t − k) = Ŝw (k), k = 1, ..Ndf t /2 − 1 (11)
∗
where Ŝw (k) is the complex conjugate of Ŝw (k). This signal is converted to the
time domain using the IDFT:
Ndf t −1
1 X
ŝl (n) = Ŝw (k)ej2πkn/Ndf t (12)
Ndf t
k=0

Where Ndf t > 2N , to support the overlap add procedure below.

We introduce the notation ŝl (n) to denote the synthesised speech for the l-th
frame. To reconstruct a continuous synthesised speech waveform, we need to
smoothly connect adjacent synthesised frames of speech. This is performed by
windowing each frame of synthesised speech, then shifting and superimposing
adjacent frames using an overlap add algorithm. A triangular window is defined
by: 
n/N,
 0≤n<N
t(n) = 1 − (n − N )/N, N ≤ n < 2N (13)

0, otherwise


The frame size, N = 80, is the same as the encoder. The shape and overlap
of the synthesis window is not important, as long as sections separated by the

11
frame size (frame to frame shift) sum to 1:

t(n) + t(N − n) = 1 (14)

The continuous synthesised speech signal ŝ(n) for the l-th frame is obtained
using:
(
ŝ(n + (l − 1)N ) + ŝl (Ndf t − N + 1 + n)t(n), n = 0, 1, ..., N − 2
ŝ(n+lN ) =
ŝl (n − N − 1)t(n) n = N − 1, .., 2N − 1
(15)
From the Ndf t samples produced by the IDFT (12), after windowing we have
2N output samples. The first N output samples n = 0, ...N − 1 complete the
current frame l and are output from the synthesiser. However we must also
compute the contribution to the next frame n = N, N + 1, ..., 2N − 1. These are
stored, and added to samples from the next synthesised frame.

3.4 Non-Linear Pitch Estimation

The Non-Linear Pitch (NLP) pitch estimator was developed by the author,
described in detail in chapter 4 of [10], and portions of this description are
reproduced here. The post processing algorithm used for pitch estimation in
Codec 2 is different from [10] and is described here. The C code nlp.c is a
useful reference for the fine details of the implementation, and the Octave script
plnlp.m can by used to plot the internal states and single step through speech,
illustrating the operation of the algorithm.
The core pitch detector is based on a square law non-linearity, that is applied
directly to the input speech signal. Given speech is composed of harmonics
separated by F0 the non-linearity generates intermodulation products at F0 ,
even if the fundamental is absent from the input signal due to high pass filtering.
Figure 7 illustrates the algorithm. The fundamental frequency F0 is esti-
mated in the range of 50-400 Hz. The algorithm is designed to take blocks of
M = 320 samples at a sample rate of 8 kHz (40 ms time window). This block
length ensures at least two pitch periods lie within the analysis window at the
lowest fundamental frequency.
The speech signal is first squared then notch filtered to remove the DC com-
ponent from the squared time domain signal. This prevents the large amplitude
DC term from interfering with the somewhat smaller amplitude term at the
fundamental. This is particularly important for male speakers, who may have
low frequency fundamentals close to DC. The notch filter is applied in the time
domain and has the experimentally derived transfer function:

1 − z −1
Hnotch (z) = (16)
1 − 0.95z −1
Before transforming the squared signal to the frequency domain, the signal
is low pass filtered and decimated by a factor of 5. This operation is performed
to limit the bandwidth of the squared signal to the approximate range of the

12
 Figure 7: The Non-Linear Pitch (NLP) algorithm

Input DC Notch Low Pass

× ↓5
Speech Filter Filter

Sub
F0 Refinement Multiple Peak Pick DFT
Search

fundamental frequency. All energy in the squared signal above 400 Hz is su-
perfluous and would lower the resolution of the frequency domain peak picking
stage. The low pass filter used for decimation is an FIR type with 48 taps and
a cut off frequency of 600 Hz. The decimated signal is then windowed and the
Ndf t = 512 point DFT power spectrum Fw (k) is computed by zero padding the
decimated signal, where k is the DFT bin.
The DFT power spectrum of the squared signal Fw (k) generally contains
several local maxima. In most cases, the global maxima will correspond to
F0 , however occasionally the global maxima |Fw (kmax )| corresponds to a spu-
rious peak or multiple of F0 . Thus it is not appropriate to simply choose the
global maxima as the fundamental estimate for this frame. Instead, we look at
submultiples of the global maxima frequency kmax /2, kmax /3, ...kmin for local
maxima. If local maxima exists and is above an experimentally derived thresh-
old we choose the submultiple as the F0 estimate. The threshold is biased down
for F0 candidates near the previous frames F0 estimate, a form of backwards
pitch tracking.
The accuracy of the pitch estimate in then refined by maximising the func-
tion:
XL
E(ω0 ) = |Sw (⌊rm⌉)|2 (17)
m=1

where r = ω0 Ndf t /2π maps the harmonic number m to a DFT bin. This
function will be maximised when mω0 aligns with the peak of each harmonic,
corresponding with an accurate pitch estimate. It is evaluated in a small range
about the coarse F0 estimate.
There is nothing particularly unique about this pitch estimator or it’s per-
formance. There are occasional artefacts in the synthesised speech that can be
traced to “gross” and “fine” pitch estimator errors. In the real world no pitch es-
timator is perfect, partially because the model assumptions around pitch break
down (e.g. in transition regions or unvoiced speech). The NLP algorithm could
benefit from additional review, tuning and better pitch tracking. However it ap-
pears sufficient for the use case of a communications quality speech codec, and
is a minor source of artefacts in the synthesised speech. Other pitch estimators

13
could also be used, provided they have practical, real world implementations
that offer comparable performance and CPU/memory requirements.

3.5 Voicing Estimation

Voicing is determined using a variation of the MBE voicing algorithm [3]. Voiced
speech consists of a harmonic series of frequency domain impulses, separated by
ω0 . When we multiply a segment of the input speech samples by the window
function w(n), we convolve the frequency domain impulses with W (k), the DFT
of the w(n). Thus for the m-th voiced harmonic, we expect to see a copy of the
window function W (k) in each band Sw(k), k = am , ..., bm . The MBE voicing
algorithm starts with the assumption that the band is voiced, and measures the
error between Sw (k) and the ideal voiced harmonic Ŝw (k).
For each band we first estimate the complex harmonic amplitude (magnitude
and phase) using [3]:
Pbm ∗
k=am Sw (k)W (k − ⌊mr⌉)
Bm = Pbm (18)
| k=a m
W (k − ⌊mr⌉)|2

where r = ω0 Ndf t /2π is a constant that maps the m-th harmonic to a DFT bin,
and ⌊x⌉ is the rounding operator. As w(n) is a real and even, W (k) is real and
even so we can write:
Pbm
k=am Sw (k)W (k + ⌊mr⌉)
Bm = P bm
(19)
2
k=am |W (k + ⌊mr⌉)|

Note this procedure is different to the Am magnitude estimation procedure in

(7), and is only used locally for the MBE voicing estimation procedure. Unlike
(7), the MBE amplitude estimation (19) assumes the energy in the band of
Sw (k) is from the DFT of a sine wave, and Bm is complex valued.
The synthesised frequency domain speech for this band is defined as:

Ŝw (k) = Bm W (k + ⌊mr⌉), k = am , ..., bm − 1 (20)

The error between the input and synthesised speech in this band is then:
m −1
bX
Em = |Sw (k) − Ŝw (k)|2
k=am
(21)
m −1
bX
2
= |Sw (k) − Bm W (k + ⌊mr⌉)|
k=am

A Signal to Noise Ratio (SNR) ratio is defined as:

m
X1000
A2m
SN R = (22)
m=1
Em

14
where m1000 = ⌊L/4⌉ is the band closest to 1000 Hz, and {Am } are computed
from (7). If the energy in the bands up to 1000 Hz is a good match to a
harmonic series of sinusoids then Ŝw (k) ≈ Sw (k) and Em will be small compared
to the energy in the band resulting in a high SNR. Voicing is declared using the
following rule: (
1, SN R > 6dB
v= (23)
0, otherwise
The voicing decision is post processed by several experimentally derived rules
to prevent common voicing errors, see the C source code in sine.c for details.

3.6 Phase Synthesis

In Codec 2 the harmonic phases {θm } are not transmitted, instead they are
synthesised at the decoder from the remaining model parameters, {Am }, ω0 ,
and v. The phase model described in this section is referred to as “zero order” or
phase0 in the source code, as it requires zero model parameters to be transmitted
over the channel.
Consider the source-filter model of speech production:

Ŝ(z) = E(z)H(z) (24)

where E(z) is an excitation signal with a relatively flat spectrum, and H(z)
is a synthesis filter that shapes the magnitude spectrum. The phase of each
harmonic is the sum of the excitation and synthesis filter phase:
h i
arg Ŝ(ejω0 m ) = arg E(ejω0 m )H(ejω0 m )
 

θ̂m = arg E(ejω0 m ) + arg H(ejω0 m )

    (25)
jω0 m
 
= ϕm + arg H(e )

For voiced speech E(z) is an impulse train (in both the time and frequency
domain). We can construct a time domain excitation pulse train using a sum
of sinusoids:
X L
e(n) = cos(mω0 (n − n0 )) (26)
m−1

Where n0 is a time shift that represents the pulse position relative to the centre
of the synthesis frame n = 0. By finding the DTCF transform of e(n) we can
determine the phase of each excitation harmonic:

ϕm = −mω0 n0 (27)

As we don’t transmit any phase information the pulse position n0 is unknown

at the decoder. Fortunately, the ear is insensitive to the absolute position of
pitch pulses in voiced speech, as long as they evolve smoothly over time (dis-
continuities in phase are a characteristic of unvoiced speech).

15
 The excitation pulses occur at a rate of ω0 (one for each pitch period). The
phase of the first harmonic advances by N ϕ1 radians over a synthesis frame of N
samples. For example if ω1 = π/20 (200 Hz), then over a (10ms N = 80) sample
frame, the phase of the first harmonic would advance (π/20)80 = 4π radians or
two complete cycles. We therefore derive n0 from the excitation phase of the
fundamental, which we treat as a timing reference. Each frame we advance the
phase of the fundamental:
ϕl1 = ϕl−1
1 + N ω0 (28)
Given ϕ1 we can compute n0 and the excitation phase of the other harmonics:

n0 = −ϕ1 /ω0
ϕm = −mω0 n0 (29)
= mϕ1 m = 2, ..., L

For unvoiced speech E(z) is a white noise signal. At each frame, we sample a
random number generator on the interval −π...π to obtain the excitation phase
of each harmonic. We set F0 = 50 Hz to use a large number of harmonics
L = 4000/50 = 80 for synthesis to best approximate a noise signal.
The second phase component is provided by sampling the phase of H(z) at
the harmonic centres. The phase spectra of H(z) is derived from the magnitude
response using minimum phase techniques. The method for deriving the phase
spectra of H(z) differs between Codec 2 modes and is described below in Sections
3.7 and 3.8. This component of the phase tends to disperse the pitch pulse
energy in time, especially around spectral peaks (formants).
The zero phase model tends to make speech with background noise sound
”clicky”. With high levels of background noise the low level inter-formant parts
of the spectrum will contain noise rather than speech harmonics, so modelling
them as voiced (i.e. a continuous, non-random phase track) is inaccurate. Some
codecs (like MBE) have a mixed voicing model that breaks the spectrum into
voiced and unvoiced regions. However (5-12) bits/frame (5-12) are required
to transmit the frequency selective voicing information. Mixed excitation also
requires accurate voicing estimation (parameter estimators always break occa-
sionally under exceptional conditions).
In our case we use a post processing approach which requires no additional
bits to be transmitted. The decoder measures the average level of the back-
ground noise during unvoiced frames. If a harmonic is less than this level it is
made unvoiced by randomising it’s phases. See the C source code for implemen-
tation details.
Comparing to speech synthesised using original phases {θm } the following
observations have been made:
1. Through headphones speech synthesised with this model drops in quality.
Through a small loudspeaker it is very close to original phases.
2. If there are voicing errors, the speech can sound clicky or staticy. If voiced
speech is mistakenly declared unvoiced, this model tends to synthesise

16
 annoying impulses or clicks, as for voiced speech H(z) is relatively flat
(broad, high frequency formants), so there is very little dispersion of the
excitation impulses through H(z).
3. When combined with amplitude modelling or quantisation, such that H(z)
is derived from {Âm } there is an additional drop in quality.

4. This synthesis model (e.g. a pulse train exciting a LPC filter) is effectively
the same as a simple LPC-10 vocoders, and yet (especially when arg[H(z)]
is derived from unquantised {Am }) sounds much better. Conventional
wisdom (AMBE, MELP) says mixed voicing is required for high quality
speech.

5. If H(z) is changing rapidly between frames, its phase contribution may

also change rapidly. This approach could cause some discontinuities in the
phase at the edge of synthesis frames, as no attempt is made to make sure
that the phase tracks are continuous (the excitation phases are continuous,
but not the final phases after filtering by H(z)).

6. The recent crop of neural vocoders produce high quality speech using
a similar parameters set, and notably without transmitting phase infor-
mation. Although many of these vocoders operate in the time domain,
this approach can be interpreted as implementing a function {θ̂m } =
F (ω0 , {Am}, v). This validates the general approach used here, and as
future work Codec 2 may benefit from being augmented by machine learn-
ing.

3.7 LPC/LSP based modes

In this and the next section we explain how the codec building blocks above
are assembled to create a fully quantised Codec 2 mode. This section discusses
the higher bit rate (3200 - 1200) modes that use a Linear Predictive Coding
(LPC) and Line Spectrum Pairs (LSPs) to quantise and transmit the spectral
magnitude information. There is a great deal of information available on these
topics so they are only briefly described here.
The source-filter model of speech production was introduced above in Equa-
tion (24). A spectrally flat excitation source E(z) excites a filter H(z) which
models the magnitude spectrum of the speech. In Linear Predictive Coding
(LPC), we define H(z) as an all-pole filter:

G G
H(z) = Pp = (30)
1− k=1 ak z −k A(z)

where {ak }, k = 1..10 is a set of p linear prediction coefficients that characterise

the filters frequency response and G is a scalar gain factor. The coefficients are
time varying and are extracted from the input speech signal, typically using a
least squares approach. An excellent reference for LPC is [7].

17
Figure 8: LPC spectrum |H(ejω )| (green line) and LSP frequencies {ωi } (green
crosses) for the speech frame in Figure 1. The original speech spectrum (blue)
and Am estimates (red) are provided as references.

80

60
Amplitude (dB)

40

20

1000 2000 3000 4000

Frequency (Hz)

To be useful in low bit rate speech coding it is necessary to quantise and

transmit the LPC coefficients using a small number of bits. Direct quantisa-
tion of these LPC coefficients is inappropriate due to their large dynamic range
(8-10 bits/coefficient). Thus for transmission purposes, especially at low bit
rates, other forms such as the Line Spectral Pair (LSP) [4] frequencies are used
to represent the LPC parameters. The LSP frequencies can be derived by de-
composing the p-th order polynomial A(z), into symmetric and anti-symmetric
polynomials P (z) and Q(z), shown here in factored form:
p/2
Y
−1
P (z) = (1 + z ) (1 − 2cos(ω2i−1 z −1 + z −2 )
i=1
(31)
p/2
Y
Q(z) = (1 − z −1 ) (1 − 2cos(ω2i z −1 + z −2 )
i=1

where ω2i−1 and ω2i are the LSP frequencies, found by evaluating the polyno-
mials on the unit circle. The LSP frequencies are interlaced with each other,
where 0 < ω1 < ω2 <, ..., < ωp < π. The separation of adjacent LSP frequen-
cies is related to the bandwidth of spectral peaks in H(z) = G/A(z). A small
separation indicates a narrow bandwidth, as shown in Figure 8. A(z) may be

18
reconstructed from P (z) and Q(z) using:
P (z) + Q(z)
A(z) = (32)
2
Thus to transmit the LPC coefficients using LSPs, we first transform the LPC
model A(z) to P (z) and Q(z) polynomial form. We then solve P (z) and Q(z)
for z = ejω to obtain p LSP frequencies {ωi }. The LSP frequencies are then
quantised and transmitted over the channel. At the receiver the quantised LSPs
are then used to reconstruct an approximation of A(z). More details on LSP
analysis can be found in [10] and many other sources.
Figure 9 presents the LPC/LSP mode encoder. Overlapping input speech
frames are processed every 10ms (N = 80 samples). LPC analysis determines
a set of p = 10 LPC coefficients {ak } that describe the spectral envelope of the
current frame and the LPC energy E = G2 . The LPC coefficients are trans-
formed to p = 10 LSP frequencies {ωi }. The source code for these algorithms is
in lpc.c and lsp.c. The LSP frequencies are then quantised to a fixed number of
bits/frame. Other parameters include the pitch ω0 , LPC energy E, and voicing
v. The quantisation and bit packing source code for each Codec 2 mode can be
found in codec2.c. Note the spectral magnitudes {Am } are not transmitted but
are still computed for use in voicing estimation (22).

Figure 9: LPC/LSP Modes Encoder

LPC LSP
Analysis Quantisation

Decimation & Bit

s(n) Window DFT
Bit Packing Stream

NLP Est Amp Est Voicing

One of the problems with quantising spectral magnitudes in sinusoidal codecs

is the time varying number of harmonic magnitudes, as L = π/ω0 , and ω0 varies
from frame to frame. As we require a fixed bit rate for our use cases, it is desir-
able to have a fixed number of parameters. Using a fixed order LPC model is a
neat solution to this problem. Another feature of LPC modelling combined with
scalar LSP quantisation is some tolerance to variations in the input frequency
response, e.g. due to microphone or anti-alias filter shape factors (see section
3.8 for more information on this issue).

19
 Some disadvantages [7] are the LPC spectrum |H(ejω )| doesn’t follow the
spectral magnitudes Am exactly, in other words is requires a non-flat excitation
spectrum to accurately model the amplitude spectrum. The slope of the LPC
spectrum near 0 and π must be 0, which means it does not track perceptually
important low frequency information well. For high pitched speakers, LPC
tends to place poles around single harmonics, rather than tracking the spectral
envelope described by {Am}. All of these problems can be observed in Figure
8. Thus exciting the LPC model by a simple, spectrally flat E(z) will result in
some errors in the reconstructed magnitude speech spectrum.
In CELP codecs these problems can be accommodated by the (high bit rate)
excitation used to construct a non-flat E(z), and some low rate codecs such as
MELP supply supplementary low frequency information to “correct” the LPC
model.
Before bit packing, the Codec 2 parameters are decimated in time. An
update rate of 20ms is used for the highest rate modes, which drops to 40ms
for Codec 2 1300, with a corresponding drop in speech quality. The number
of bits used to quantise the LPC model via LSPs is also reduced in the lower
bit rate modes. This has the effect of making the speech less intelligible, and
can introduce annoying buzzy or clicky artefacts into the synthesised speech.
Lower fidelity spectral magnitude quantisation also results in more noticeable
artefacts from phase synthesis. Nevertheless at 1300 bits/s the speech quality
is quite usable for HF digital voice, and at 3200 bits/s comparable to closed
source codecs at the same bit rate.

Figure 10: LPC/LSP Modes Decoder

Bit
Unpack Interpolate LSP to LPC Sample Am
Stream

Phase
Post Filter
Synthesis

Sinusoidal
ŝ(n)
Synthesis

Figure 10 shows the LPC/LSP mode decoder. Frames of bits received at the
frame rate are unpacked and resampled to the 10ms internal frame rate using
linear interpolation. The spectral magnitude information is resampled by linear
interpolation of the LSP frequencies, and converted back to a quantised LPC
model Ĥ(z). The harmonic magnitudes are recovered by averaging the energy

20
of the LPC spectrum over the region of each harmonic:
v
u bm −1
uX
Âm = t |Ĥ(k)|2 (33)
k=am

where Ĥ(k) is the Ndf t point DFT of the received LPC model for this frame.
For phase synthesis, the arg[H(z)] component is determined by sampling Ĥ(k)
in the centre of each harmonic:
h i
arg H(ejω0 m ) = arg Ĥ(⌊mr⌉)
 
(34)

Figure 11: LPC post filter. LPC spectrum before |H(ejω )| (green line) and after
(red) post filtering. The distance between the spectral peaks and troughs has
been increased. The step change at 1000 Hz is +3dB low frequency boost (see
source code).

80

60
Amplitude (dB)

40

20

1000 2000 3000 4000

Frequency (Hz)

Prior to sampling the amplitude and phase, a frequency domain post filter
is applied to the LPC power spectrum. The algorithm is based on the MBE
frequency domain post filter [6, Section 8.6, p 267], which is in turn based on
the frequency domain post filter from McAulay and Quatieri [5, Section 4.3, p
148]. The authors report a significant improvement in speech quality from the
post filter, which has also been our experience when applied to Codec 2. The
post filter is given by:
Pf (ejω ) = g Rw (ejω )β


(35)
Rw (jω ) = A(ejω/γ )/A(ejω )

21
where g is chosen to normalise the gain of the post filter, and β = 0.2, γ = 0.5
are experimentally derived constants. The post filter raises the spectral peaks
(formants), and lowers the inter-formant energy. The γ term compensates for
spectral tilt, providing equal emphasis at low and high frequencies. The authors
suggest the post filter reduces the noise level between formants, an explanation
commonly given to post filters used for CELP codecs where significant inter-
formant noise exists from the noisy excitation source. However, in harmonic
sinusoidal codecs, there is no excitation noise between formants in E(z). Our
theory is the post filter also acts to reduce the bandwidth of spectral peaks,
modifying the energy distribution across the time domain pitch cycle which
improves speech quality, especially for low pitched speakers.
A disadvantage of the post filter is the need for experimentally derived con-
stants. It performs a non-linear operation on the speech spectrum, and if mis-
applied can worsen speech quality. As it’s operation is not completely under-
stood, it represents a source of future quality improvement.

3.8 Codec 2 700C

To efficiently transmit spectral amplitude information, Codec 2 700C uses a set
of algorithms collectively denoted newamp1. One of these algorithms is the Rate
K resampler which transforms the variable length vectors of spectral magnitude
samples to fixed length K vectors suitable for vector quantisation. Figure 12
presents the Codec 2 700C encoder.

Figure 12: Codec 2 700C (newamp1) Encoder

s(n) Window DFT Est Amp

sw (n) Sω (k)
a

Microphone Resample
NLP
EQ b Rate K

Decimate Est
log ω0
& VQ Voicing

Bit Packing Bit Stream

22
 Consider a vector a of L harmonic spectral magnitudes expressed in dB:
 
a = 20log10 A1 , 20log10 A2 , . . . 20log10 AL (36)
   
Fs π
L= = (37)
2F0 ω0
F0 and L are time varying as the pitch track evolves over time. For speech
sampled at Fs = 8 kHz F0 is typically in the range of 50 to 400 Hz, giving L in
the range of 10 . . . 80.
To quantise and transmit a, it is convenient to resample a to a fixed length
K element vector b using a resampling function:
 
b = B1 , B2 , . . . BK = R(a) (38)

Where R is a resampling function. To model the response of the human ear Bk

are sampled on K non-linearly spaced points on the frequency axis:

fk = warp(k, K) Hz k = 1...K
warp(1, K) = 200 Hz (39)
warp(K, K) = 3700 Hz

where warp() is a frequency warping function. Codec 2 700C uses K = 20, and
warp() is defined using the Mel function [9, p 150] (Figure 13) which samples the
spectrum more densely at low frequencies, and less densely at high frequencies:

mel(f ) = 2595log10 (1 + f /700) (40)

The inverse mapping of f in Hz from mel(f ) is given by:

f = mel−1 (x) = 700(10x/2595 − 1); (41)

We wish to use mel(f ) to construct warp(k, K), such that there are K
evenly spaced points on the mel(f ) axis (Figure 14). Solving for the equation of
a straight line we can obtain mel(f ) as a function of k, and hence warp(k, K)
(Figure 15):
mel(3700) − mel(200)
g=
K −1 (42)
mel(f ) = g(k − 1) + mel(200)

where g is the gradient of the line. Substituting (41) into the LHS:

2595log10 (1 + f /700) = g(k − 1) + mel(200)

(43)
fk = warp(k, K) = mel−1 (g(k − 1) + mel(200))

and the inverse warp function:

mel(f ) − mel(200)
k = warp−1 (f, K) = +1 (44)
g

23
 Figure 13: Mel function

The input speech may be subject to arbitrary filtering, for example, due
to the microphone frequency response, room acoustics, and anti-aliasing filter.
This filtering is fixed or slowly time-varying. The filtering biases the target
vectors away from the VQ training material, resulting in significant additional
mean square error. The filtering does not greatly affect the input speech quality,
however the VQ performance distortion increases and the output speech quality
is reduced. This is exacerbated by operating in the log domain, the VQ will try
to match very low level, perceptually insignificant energy near 0 and 4000 Hz. A
microphone equaliser algorithm has been developed to help adjust to arbitrary
microphone filtering.
For every input frame l, the equaliser (EQ) updates the dimension K equaliser
vector e:
el = el−1 + β(b − t) (45)
where t is a fixed target vector set to the mean of the VQ quantiser, and β is a
small adaption constant.
The equalised, mean removed rate K vector d is vector quantised for trans-

24
 Figure 14: Linear mapping of mel(f ) to Rate K sample index k

mel(f)

(K,mel(3700))

(1,mel(200))

mission over the channel:

c=b−e
d = c − c̄
(46)
ĉ = V Q(d) + Q(c̄)
ˆ
= d̂ + c̄

Codec 2 700C uses a two stage VQ with 9 bits (512 entries) per stage. The mbest
multi-stage search algorithm is used to jointly search the two stages (using 5
survivors from the first stage). Note that VQ is performed in the log amplitude
(dB) domain. The mean of c is removed prior to VQ and scalar quantised and
transmitted separately as the frame energy. At the decoder, the rate L vector
â can then be recovered by resampling â:

â = S(ĉ + p) (47)

where p is a post filter vector. The post filter vector is generated from the
mean-removed rate K vector d̂ in the log frequency domain:
 
p = G + Pgain d̂ + r − r

r = R1 , R2 , . . . RK
 (48)
Rk = 20log10 (fk /300) k = 1, ..., K

where G is an energy normalisation term, and 1.2 < Pgain < 1.5 describes the
amount if post filtering applied. G and Pgain are similar to g and β in the
LPC/LSP post filter (35). The r term is a high pass (pre-emphasis) filter with
+20 dB/decade gain after 300 Hz (fk is given in (43)). The post filtering is
applied on the pre-emphasised vector, then the pre-emphasis is removed from
the final result. Multiplying by Pgain in the log domain is similar to the α power

25
 Figure 15: warp(k, K) function for K = 20

function in (35); spectral peaks are moved up, and troughs pushed down. This
filter enhances the speech quality but also introduces some artefacts.
Figure 16 is the block diagram of the decoder signal processing. Cepstral
techniques are used to synthesise a phase spectra arg[H(ejω ]) from â using a
minimum phase model.
Some notes on the Codec 2 700C newamp1 algorithms:
1. The amplitudes and Vector Quantiser (VQ) entries are in dB, which
matches the ear’s logarithmic amplitude response.
2. The mode is capable of communications quality speech and is in common
use with FreeDV, but is close to the lower limits of intelligibility, and
doesn’t do well in some languages (problems have been reported with
German and Japanese).

3. The VQ was trained on just 120 seconds of data - way too short.
4. The parameter set (pitch, voicing, log spectral magnitudes) is very similar
to that used for the latest neural vocoders.
5. The Rate K algorithms were recently revisited, and several improvements
were proposed and prototyped [2].

26
 Figure 16: Codec 2 700C (newamp1) Decoder

Bit ĉ
Unpack Interpolate Post Filter
Stream

ĉ + p

Resample
ωˆ0 , v
to Rate L

â

Phase Sinusoidal
ŝ(n)
Synthesis Synthesis

4 Summary of Codec 2 Modes

Mode Frm Bits Am E ω0 v Use Cases

(ms)
3200 20 64 50 5 7 2 M17
2400 20 50 36 8 - 2
1600 40 64 36 10 14 4 M17
1400 40 56 36 16 - 4
1300 40 52 36 5 7 4 FreeDV 1600
1200 40 48 27 16 - 4
700C 40 28 18 4 6 - FreeDV 700C/D/E

Table 2: Codec 2 Modes

The 3200 mode quantises the LSP differences ωi+1 − ωi , which provides low
distortion at the expense of robustness to bit errors, as an error in a low order
LSP difference will propagate through the frame. The 2400 and 1200 bit/s
modes use a joint delta ω0 and energy VQ, which is efficient but also suffers
from error propagation so is not suitable for high BER use cases.
There is an unfortunate overlap in the naming conventions of Codec 2 and
FreeDV. The Codec 2 700C mode is used in the FreeDV 700C, 700D, and 700E
modes.

5 Summary of Codec 2 Source Files

Codec 2 is part of the codec2 repository, which also includes various modems and
FreeDV API code. This section lists the files specific to the speech codec. The

27
cmake system builds the libcodec2 library, which is called by user applications
via the Codec 2 API in codec2.h. See the repository README for information
on building, demo applications, and an introduction to other features of the
codec2 repository.

File Description
c2dec Sample decoder application
c2enc Sample encoder application
c2sim Simulation and development application
codebook Directory containing quantiser tables
codec2.c Quantised encoder and decoder functions that implement each mode
codec2 fft.c Wrapper for FFT (usually kiss FFT)
defines.h Constants
lpc.c LPC functions
mbest.c Multistage VQ search
newamp1.c Codec 2 700C newamp1 mode
nlp.c Non-linear Pitch (NLP)
sine.c Sinusoidal analysis, synthesis, voicing estimation
phase.c Phase synthesis
quantise.c Quantisation, in particular for LPC/LSP modes

Table 3: Codec 2 Source Files

6 Glossary

Acronym Description
DFT Discrete Fourier Transform
DTCF Discrete Time Continuous Frequency Fourier Transform
EQ (microphone) Equaliser
IDFT Inverse Discrete Fourier Transform
LPC Linear Predictive Coding
LSP Line Spectrum Pair
MBE Multi-Band Excitation
MSE Mean Square Error
NLP Non Linear Pitch (algorithm)
VQ Vector Quantiser

Table 4: Glossary of Acronyms

28
 Symbol Description Units
A(z) LPC (analysis) filter
am Lower DFT index of current band
bm Upper DFT index of current band
{Am } Set of harmonic magnitudes m = 1, ...L dB
a {Am } in vector form
Bm Complex spectral amplitudes used for voicing estimation
E Frame energy
E(z) Excitation in source-filter model
F0 Fundamental frequency (pitch) Hz
Fs Sample rate (usually 8 kHz) Hz
Fw (k) DFT of squared speech signal in NLP pitch estimator
G LPC gain
H(z) Synthesis filter in source-filter model
Ĥ(z) Synthesis filter approximation after quantisation
l Frame index
L Number of harmonics
N Processing frame size in samples
n0 Excitation pulse position
P Pitch period ms or samples
P (z), Q(z) LSP polynomials
Pf (ejω ) LPC post filter
{θm } Set of harmonic phases m = 1, ...L dB
r Maps a harmonic number m to a DFT index
s(n) Input time domain speech
ŝ(n) Output (synthesised) time domain speech
sw (n) Time domain windowed input speech
Sw (k) Frequency domain windowed input speech
Ŝw (k) Frequency domain output (synthesised)speech
t(n) Triangular synthesis window
ϕm Phase of excitation harmonic
ω0 Fundamental frequency (pitch) radians/sample
{ωi } Set of LSP frequencies
w(n) Window function
W (k) DFT of window function
v Voicing decision for the current frame

Table 5: Glossary of Symbols

7 Further Documentation Work

This section contains ideas for expanding the documentation of Codec 2. Please
contact the authors if you are interested in this material or would like to help
develop it.

29
 1. The c2sim utility is presently undocumented. We could add some worked
examples aimed at the experimenter - e.g. using c2sim to extract and
plot model parameters. Demonstrate how to listen to various stages of
quantisation.
2. Several GNU Octave scripts exist that were used to develop Codec 2. We
could add information describing how to use the Octave tools to single
step through the codec operation.

References
[1] Enhancing HF Digital Voice with FreeDV, 2023.
https://www.ardc.net/apply/grants/2023-grants/
enhancing-hf-digital-voice-with-freedv/.
[2] FreeDV-015 Codec 2 Rate K Resampler, 2023. https://github.com/
drowe67/misc/blob/master/ratek_resampler/ratek_resampler.pdf.
[3] Daniel W Griffin and Jae S Lim. Multiband excitation vocoder. IEEE
Transactions on acoustics, speech, and signal processing, 36(8):1223–1235,
1988.
[4] Fumitada Itakura. Line spectrum representation of linear predictor coeffi-
cients of speech signals. The Journal of the Acoustical Society of America,
57(S1):S35–S35, 1975.

[5] W Bastiaan Kleijn and Kuldip K Paliwal. Speech coding and synthesis.
Elsevier Science Inc., 1995.
[6] Ahmet M Kondoz. Digital speech: coding for low bit rate communication
systems. John Wiley & Sons, 1994.
[7] John Makhoul. Linear prediction: A tutorial review. Proceedings of the
IEEE, 63(4):561–580, 1975.
[8] Robert McAulay and Thomas Quatieri. Speech analysis/synthesis based
on a sinusoidal representation. IEEE Transactions on Acoustics, Speech,
and Signal Processing, 34(4):744–754, 1986.

[9] Douglas O‘Shaughnessy. Speech Communication - Human and machine.

Addison-Wesley Publishing Company, 1997.
[10] David Grant Rowe. Techniques for harmonic sinusoidal coding. Citeseer,
1997. https://www.rowetel.com/downloads/1997_rowe_phd_thesis.
pdf.

30