This Annex describes an alternative implementation of G.729 Annex A based on floating‑point arithmetic. Subjective quality tests have been performed by NTT (Japan) and CNET (France) to assess the quality of these floating‑point versions under various conditions (input level, error, background noise, tandeming). Different interoperability configurations with the fixed‑point version of the algorithm have also been tested. These tests proved full interoperability of this floating‑point implementation to both Recommendation G.729 and its Annex A. The design of a set of test vectors remains for further study.