Unified and Optimal Unit-Selection Framework

  • V. Ramasubramanian
  • Harish Doddala
Part of the SpringerBriefs in Electrical and Computer Engineering book series


This chapter is devoted to the class of unit-selection algorithms proposed by us earlier, and which represent an optimal and unified generalization over the single-frame and sub-optimal segmental unit-selection algorithms of Lee and Cox dealt with in the previous chapter. Following a detailed treatment of this unified formulation, we present the modified one-pass DP algorithm which provides an optimal solution to this unified formulation, and characterize its performance in terms of rate-distortion curves obtained using large unit-database size. We first show how it generalizes over the single-frame algorithm towards handling longer fixed length units, with progressively enhanced rate-distortion performance with increase in unit-size. Subsequently, we compare the rate-distortion performance of the proposed algorithm with the sub-optimal segmental algorithm of Lee and Cox and demonstrate the clear advantages accrued. Following this, we answer the question of what, if any, is the advantage of moving from small clustered segment codebook, as in the classic Shiraki and Honda’s variable length quantization algorithm, to large unit-database sizes as in the unit-selection framework, and through rate-distortion performances, show the highly enhanced performances of the unit-selection algorithms in comparison to the conventional vector quantization, matrix quantization and variable length segment quantization using clustered codebooks.


  1. [HR08]
    D. Harish, V. Ramasubramanian, Comparison of segment quantizers: VQ, MQ, VLSQ and Unit-selection algorithms for ultra low bit-rate speech coding, in Proceedings of ICASSP ’08, Las Vegas, Mar 2008, pp. 4773–4776Google Scholar
  2. [LC01]
    K.S. Lee, R.V. Cox, A very low bit rate speech coder based on a recognition/synthesis paradigm. IEEE Trans. Speech Audio Process. 9(5), 482–491 (2001)CrossRefGoogle Scholar
  3. [LC02]
    K.S. Lee, R.V. Cox, A segmental speech coder based on a concatenative TTS. Speech Commun. 38, 89–100 (2002)MATHCrossRefGoogle Scholar
  4. [N84]
    H. Ney, The use of one-stage dynamic programming algorithm for connected word recognition. IEEE Trans. Acoust. Speech Signal Process. 32(2), 263–271 (1984)CrossRefGoogle Scholar
  5. [RH07]
    V. Ramasubramanian, D. Harish, An optimal unit-selection algorithm for ultra low bit-rate speech coding, in Proceedings of ICASSP ’07, Hawaii, Apr 2007, pp. IC-541–IC-544Google Scholar
  6. [RH06]
    V. Ramasubramanian, D. Harish, An unified unit-selection framework for ultra low bit-rate speech coding, in Proceedings of Interspeech ’06, Pittsburgh, Sept 2006, pp. 217–220Google Scholar
  7. [SH88]
    Y. Shiraki, M. Honda, LPC speech coding based on variable-length segment quantization”. IEEE Trans. Acoust. Speech Signal Process. 36(9), 1437–1444 (1988)MATHCrossRefGoogle Scholar
  8. [TG85]
    C. Tsao, R.M. Gray, Matrix quantizer design for LPC speech using the generalized Lloyd algorithm. IEEE Trans. ASSP 33(3), 537–545 (1985)CrossRefGoogle Scholar
  9. [W82]
    D.Y. Wong et al., An 800 b/s vector quantization LPC vocoder. IEEE Trans. ASSP 30(6), 770–780 (1982)CrossRefGoogle Scholar

Copyright information

© The Author 2015

Authors and Affiliations

  • V. Ramasubramanian
    • 1
  • Harish Doddala
    • 2
  1. 1.PES Institute of Technology – Bangalore South CampusBangaloreIndia
  2. 2.OracleRedwood ShoresUSA

Personalised recommendations