Organisation internationale de normalisation

Lossless Coding of Pitch Lag

Takehiro Moriya, NTT, presented new information on lossless pitch lag coding. Specifically, he presented

• 
  The dynamics of selection of ACELP coding mode (i.e. number of pulses used) at 12 kb/s. He noted that jumping to a higher ACELP coding mode involved using some of the “saved bits” from the CE but also some bits from the general bit reservoir. In other words, half of the enhancement comes from pitch enhancement and half from better rate control.
  
• 
  Sequence of ACELP and TCX mode selection: out of 949 superframes, 816 are identical (RM8 vs CE) and 133 are different.
  
• 
  The Huffman table was trained using 128 k ACELP frames: 50% are USAC reference bitstreams and 50% are material outside the USAC test set. The presenter noted that the Huffman table’s degrees of freedom is less than 64.
  

Max Neuendorf, FhG, asked about the Huffman codebook training database. Specifically, he expressed a concern that the test set was used in the training database and that, for another portion of the database, only a subset of the operating modes were used.

Heiko Purnhagen, Dolby and Philippe Gournay, VoiceAge, noted that it is not clear what portion of the increase in performance comes from the bit savings and what portion comes from shifting the bit allocation from TCX to ACELP.

Bernhard Grill, FhG, noted that FhG could make a large database of bitstreams available to re-train the CE’s Huffman codebook.

This discussion will be continued in Friday’s Audio plenary.