Communications Fundamentals FAQ / Chapter 3

3.8.10 What Is the Difference Between Waveform, Model-Based, and Hybrid Speech Coding?

1. Why Is Speech Coding Important?
2. What Is a Speech Coder?
3. What Is Waveform Coding?
4. How Do Waveform Coders Work?
5. What Are Examples of Waveform Coders?
6. What Are the Advantages of Waveform Coding?
7. What Are the Disadvantages of Waveform Coding?
8. What Is Model-Based Coding?
9. How Is Human Speech Produced?
10. What Is a Vocoder?
11. What Is Linear Predictive Coding (LPC)?
12. What Are the Advantages of Model-Based Coding?
13. What Are the Disadvantages of Model-Based Coding?
14. What Is Hybrid Coding?
15. How Do Hybrid Coders Work?
16. What Is CELP?
17. Why Is CELP So Successful?
18. Where Are Hybrid Coders Used?
19. How Do the Three Approaches Compare?
20. Which Approach Is Used Today?
21. Why Is Understanding These Approaches Important?

Speech is one of the most important sources of information carried by communications systems. From traditional telephone networks and mobile phones to satellite systems, military radios, and Internet communications, enormous effort has been devoted to representing speech efficiently while maintaining acceptable quality.

At first glance, speech coding appears to be a straightforward problem. After all, speech can be sampled and converted into digital form using pulse-code modulation (PCM). However, PCM requires 64 kbps for a standard telephone-quality voice channel, and engineers soon sought methods that could reduce this bit rate while preserving intelligibility and naturalness.

Over time, three major approaches to speech coding emerged:

• Waveform coding.
• Model-based coding.
• Hybrid coding.

Each approach reflects a different philosophy regarding how speech should be represented. Some attempt to reproduce the waveform accurately. Others attempt to model how speech is generated. Still others combine elements of both approaches.

Understanding these three categories provides valuable insight into the evolution of speech communications and the operation of modern voice systems.

Why Is Speech Coding Important?

Speech occupies a unique position in communications engineering.

Humans are remarkably skilled at understanding speech, even when it is distorted, noisy, or incomplete. This characteristic creates opportunities for compression because, unlike computer data, speech does not necessarily require exact reconstruction. Instead, the objective is usually to produce speech that sounds natural and remains intelligible to listeners.

By exploiting characteristics of human speech and hearing, engineers can often reduce bit rates dramatically while maintaining acceptable quality.

This has enabled:

• Digital telephony.
• Mobile communications.
• Satellite voice systems.
• Voice over IP.
• Military communications.
• Voice messaging services.

What Is a Speech Coder?

A speech coder, or vocoder (voice coder), is a source coder designed specifically for speech signals.

Its purpose is to convert speech into a digital representation requiring fewer bits than direct PCM transmission. The speech coder analyzes the incoming speech signal and produces a compressed bit stream. At the receiver, a decoder reconstructs an approximation of the original speech.

The exact method depends on the coding approach employed.

What Is Waveform Coding?

Waveform coding attempts to preserve the shape of the original speech waveform.

The coder does not attempt to understand how speech is produced. Instead, it simply seeks to reproduce the signal as accurately as possible. The philosophy is straightforward: If the waveform is reproduced accurately, the speech should sound natural.

Waveform coders therefore focus on signal fidelity rather than speech-production mechanisms.

How Do Waveform Coders Work?

Waveform coders analyze the speech signal directly.

Typical techniques include:

• Sampling.
• Quantization.
• Prediction.
• Difference encoding.

The objective is to minimize the difference between:

• The original waveform.
• The reconstructed waveform.

The resulting speech typically sounds natural because the waveform closely resembles the original signal.

What Are Examples of Waveform Coders?

Several important waveform coders have been developed:

• Pulse-Code Modulation (PCM). PCM is the simplest and most familiar waveform coder. Standard telephone PCM uses 8 kHz sampling; 8 bits per sample; producing a data rate of 64 kbps.
• Differential PCM (DPCM). Rather than transmitting absolute sample values, DPCM transmits differences between successive samples. This can reduce bit rate while maintaining quality.
• Adaptive Differential PCM (ADPCM). ADPCM improves upon DPCM by adapting its parameters to changing signal conditions. Bit rates of: 32 kbps or lower are common.

What Are the Advantages of Waveform Coding?

Waveform coders offer several important advantages.

• Natural sounding speech. The reconstructed waveform closely resembles the original signal.
• Good quality. Speech generally sounds realistic and familiar.
• General signal representation. Waveform coders can often handle speech, music, tones, and other audio signals. without significant modification.
• Relative simplicity. Many waveform coders are straightforward to implement.

What Are the Disadvantages of Waveform Coding?

The principal disadvantage is bit rate.

Because waveform coders attempt to reproduce the entire signal accurately, they typically require relatively large amounts of data (typically 64 kbps for PCM compared with 16-32 kbps for ADPCM.

These rates may be excessive in bandwidth-constrained systems.

What Is Model-Based Coding?

Model-based coding takes a completely different approach.

Rather than attempting to reproduce the waveform itself, model-based coders attempt to reproduce the process that generates speech. The underlying idea is that human speech is produced by a physical mechanism and, if that mechanism can be modeled mathematically, only the model parameters need to be transmitted.

This can dramatically reduce bit rate.

How Is Human Speech Produced?

Speech originates within the vocal system.

The major components include:

• Lungs.
• Vocal cords.
• Vocal tract.
• Tongue.
• Lips.

The lungs provide airflow. The vocal cords generate excitation. The vocal tract shapes the resulting sound. Different speech sounds arise from different combinations of these elements.

Model-based coders attempt to describe this process mathematically.

What Is a Vocoder?

A vocoder is a model-based speech coder.

Instead of transmitting waveform samples, the vocoder transmits parameters describing:

• Pitch.
• Vocal-tract resonances.
• Excitation characteristics.
• Speech energy.

The receiver then synthesizes speech using these parameters.

The reconstructed speech may not reproduce the original waveform accurately, but it can remain highly intelligible.

What Is Linear Predictive Coding (LPC)?

One of the most important model-based techniques is linear predictive coding (LPC).

LPC models the vocal tract as a digital filter so that the encoder estimates:

• Filter coefficients.
• Pitch.
• Gain.
• Voicing information.

These parameters are transmitted to the receiver. The decoder reconstructs speech by exciting a corresponding digital filter.

LPC became one of the most influential developments in speech coding and remains widely used today.

What Are the Advantages of Model-Based Coding?

Model-based coders can achieve extremely low bit rates. Examples include:

This represents a dramatic reduction compared with PCM.

Such low bit rates are especially valuable in:

• Satellite systems.
• Military communications.
• HF radio systems.
• Secure communications.

What Are the Disadvantages of Model-Based Coding?

The principal disadvantage is speech quality.

Because the waveform is not reproduced accurately:

• Speech may sound synthetic.
• Naturalness may be reduced.
• Speaker characteristics may be altered.

Listeners often describe traditional vocoder speech as robotic or mechanical.

Although intelligibility may remain good, subjective quality is generally lower than that achieved by waveform coding.

What Is Hybrid Coding?

Hybrid coding emerged as an attempt to combine the strengths of both approaches.

Hybrid coders seek to achieve:

• The quality of waveform coding.
• The efficiency of model-based coding.

Rather than reproducing the waveform directly or relying entirely on a speech-production model, hybrid coders employ elements of both.

This approach has proven remarkably successful.

How Do Hybrid Coders Work?

Hybrid coders typically:

• Employ a speech-production model.
• Analyze the speech waveform.
• Select an excitation signal that minimizes reconstruction error.
• Transmit model parameters and excitation information.

The decoder uses these components to reconstruct the speech.

The result often sounds significantly more natural than traditional vocoder speech while maintaining low bit rates.

What Is CELP?

The most important family of hybrid coders is code-excited linear prediction (CELP).

CELP combines:

• Linear predictive modeling.
• Excitation codebooks.
• Perceptual optimization.

Rather than transmitting the excitation directly, the encoder selects entries from a predefined codebook. The decoder uses the same codebook to reconstruct the speech.

CELP and its descendants dominate modern speech communications.

Why Is CELP So Successful?

CELP achieves an excellent balance between quality and efficiency.

Typical bit rates include:

At these rates, speech quality can approach that of much higher-rate waveform coders.

This combination of performance and efficiency explains its widespread adoption.

Where Are Hybrid Coders Used?

Hybrid coding forms the basis of many modern systems.

• Cellular networks. Most mobile voice services employ hybrid coding.
• Voice over IP. Internet telephony relies heavily on CELP-derived codecs.
• Satellite communications. Bandwidth-efficient voice transmission often uses hybrid coders.
• Military communications. Secure voice systems frequently employ low-rate hybrid coding.
• Video conferencing. Many conferencing systems incorporate hybrid speech coding.

How Do the Three Approaches Compare?

The differences can be summarized as follows:

Each approach occupies a useful position within the speech-coding landscape.

Which Approach Is Used Today?

Modern communications systems overwhelmingly favor hybrid coding because it provides:

• Good speech quality.
• Efficient bandwidth usage.
• Acceptable computational complexity.

Waveform coding remains important where maximum quality is required. Model-based coding remains valuable where extremely low bit rates are essential.

However, for most commercial voice communications systems, hybrid coding provides the most attractive compromise.

Why Is Understanding These Approaches Important?

The evolution from waveform coding to model-based coding and finally hybrid coding illustrates a recurring theme in communications engineering:

Better performance is often achieved by exploiting knowledge of the information source. By understanding how speech is generated and perceived, engineers have reduced voice transmission rates from 64 kbps PCM channels to only a few kilobits per second while maintaining acceptable quality.

This achievement has played a major role in the development of modern mobile, satellite, and Internet communications systems.

Summary

Speech coders are commonly divided into three categories: waveform coders, model-based coders, and hybrid coders. Waveform coders such as PCM attempt to reproduce the speech waveform accurately and provide high quality but require relatively high bit rates. Model-based coders such as LPC vocoders transmit parameters describing the speech-production process and can achieve extremely low bit rates, although speech may sound synthetic.

Hybrid coders combine elements of both approaches, using speech-production models together with waveform analysis to achieve an effective balance between quality and efficiency. Modern communications systems, including mobile networks and Voice over IP services, rely heavily on hybrid coding techniques such as CELP, making them the dominant form of speech coding in use today.

Return to Chapter 3 FAQ 3.8.10