Selectable Mode Vocoder (SMV) is variable bitratespeech coding standard used in CDMA2000 networks.[1] SMV provides multiple modes of operation that are selected based on input speech characteristics.
Technical specification
Codecs
The SMV for WidebandCDMA is based on 4 codecs: full rate at 8.5kbit/s, half rate at 4kbit/s, quarter rate at 2kbit/s, and eighth rate at 800bit/s.[1] The full rate and half rate are based on the CELPalgorithm[1] that is based on a combined closed-loop-open-loop-analysis (COLA). In SMV the signal frames are first classified as:
Silence/Background noise
Non-stationary unvoiced
Stationary unvoiced
Onset
Non-stationary voiced
Stationary voiced
Algorithm
The algorithm includes voice activity detection (VAD) followed by an elaborate frame classification scheme. Silence/background noise and stationary unvoiced frames are represented by spectrum-modulated noise and coded at 1/4 or 1/8 rate. The SMV uses 4 subframes for full rate and two/three subframes for half rate. The stochastic (fixed) codebook structure is also elaborate and uses sub-codebooks each tuned for a particular type of speech. The sub-codebooks have different degrees of pulse sparseness (more sparse for noise like excitation). SMV scores a high of 3.6 MOS[2] at full rate with clean speech.
The coder works on a frame of 160 speech samples (20ms) and requires a look ahead of 80 samples (10ms) if noise-suppression option B is used. An additional 24 samples of look ahead is required if noise-suppression option A is used. So the algorithmic delay for the coder is 30ms with noise-suppression option B and 33ms with noise-suppression option A.
The next evolution of CDMA speech codecs is VMR-WB which provides much higher speech quality with wideband while fitting to the same networks.
SMV can be also used in 3GPP2 container file format – 3G2.
