There are two major groups of codecs used for Packet Telephony (such as Voice-over-Internet-Protocol technologies). The main differentiation between these two groups of codecs is the coding/decoding mechanism: in the first group there are codecs whose encoding/decoding principle does not depend on the type of encoded/decoded signals. In the second group there are codecs whose encoding/decoding mechanism is based on modeling production apparatus and process of human speech. The speech modeling codecs do not operate exceptionally well on non-speech signals (such as harmonic tones and music). Group I codecs do not have this limitations. However, they occupy greater bit-rate than Group II codecs.
An example of a codec belonging to Group I is G.711 PCM codec. Codecs belonging to Group I are called waveform codecs. They are mostly lossy codecs (meaning that the encoding/decoding operation results in some loss of information). An exception is G.711 PCM codec which a pseudo-lossless codec (a two-pass through the cycle of encoding/decoding results in ceasing the information loss).
An example of a codec belonging to Group II is G.729A codec, a Linear Prediction-type of codec; specifically, it is a CS-ACELP (Conjugate-Structure Algebraic-Code-Excited Linear-Prediction codec), one of the ITU-T-standard low-bit-rate codecs. All codecs belonging to Group II are lossy.
In Table 1, there are codecs belonging to Group I (waveform codecs) with notes on basic characteristics (cf. Ref. ).
Table 1: Common waveform codecs used in VoIP
Frame size [ms]
|G.711 (ITU-T)||PCM||64||8||Sample-based||A-law & μ-law|
|G.721 (ITU-T)||ADPCM||32||8||Sample-based||Now G.726|
|G.722 (ITU-T)||2-band ADPCM||64||16||Sample-based|
|G.722.2 (ITU-T)||AMR-WB||6.6 – 23.85||16||20||Main app: 3rd gen mobile|
|G.723 (ITU-T)||ADPCM-extension||24/40||8||Sample-based||Now G.726|
|G.726 (ITU-T)||ADPCM||16/24/32/40||8||Sample-based||Replaces G.721 and G.723|
|G.727 (ITU-T)||ADPCM||Variable||8||Sample-based||Similar to G.726|
|DVI (IMA)||DVI4 uses ADPCM||32||Variable||Sample-based|
|L16||Linear PCM||128||Variable||Sample-based||WB (uncompressed)|
In Table 2, there are codecs belonging to Group II (model-based codecs, i.e., related directly or indirectly to Linear Prediction models) with notes on basic characteristics.
Table 2: Common model-based (LP-related) codecs used in VoIP
Frame size [ms]
|G.723.1 (ITU-T)||MP-MLQ and ACELP||5.6/6.3||8||30||A part of H.323 video conf|
|G.728 (ITU-T)||LD-CELP||16||8||2.5||Very low delay|
|G.729 (ITU-T)||CS-ACELP||8||8||10||Low delay (15 ms)|
|G.729.1 (ITU-T)||CS-ACELP||8-32, step 2||8||10||WB|
|GSM 06.10 (ETSI)||RPE-LTP||13||8||22.5||Mostly in GSM cellular; seldom in VoIP|
|LPC10 (USA Gov)||LP||2.4||8||22.5||Very seldom used in VoIP|
|Speex||CELP||8,16,32||NB & WB||30 (NB), 34 (WB)||Also includes ECAN and NR features|
|DoD CELP (DoD)||CELP||4.8||30|
|EVRC (3GPP2)||RCELP||9.6/4.8/1.2||8||20||Enhanced Variable Rate CODEC; for CDMA|
|SILK (Skype)||LPC||6-40||Variable||20||Adaptive bit-rate codec; low delay|
|iLBC||AAC||8||13.3||30||NB (there is a 15,2 kbps version)|
Tables 1 and 2 include examples of codecs that are relatively frequently used in VoIP technology. The full list of waveform-codecs and model-based (for example, CELP) codecs is much longer. There are also many proprietary solutions for encoding and decoding speech signals following waveform coding as well as the model-based coding.
VOCAL’s VoIP Software modules are part of our VoIP Reference Design and provide secure, real-time unified communications for voice, video, radio and data over the Internet or any other IP network. Speech codecs are part of the software library modules. Contact us to discuss your speech application requirements with our engineering staff.
- The RCELP speech-coding algorithm, Transactions on Emerging Telecommunication Technologies; autors: W. Bastiaan Kleijn1, Peter Kroon1 and Dror Nahumi2