According to some old notes I have, V.34 uses signal rates up to 3429 baud and I don't think that V.34bis added to the modulation rates. Since 'Plain Old Telephone Service' connections are sampled at 8000 Hz for transport through the digital public switched telephone network, the Nyquist-Shannon sampling theorem states that the highest theoretical signal rate is 4000 Hz, but real-world implementations can only approach that.
That's true of every popular modulation since Bell 103J; if I recall correctly, Bell 212A used phase shifts to encode 2 bits at 600 baud for 1200 bps, V.22bis used phase and amplitude variations to encode 4 bits at 600 baud for 2400 bps, and Telebit PEP modulation used pashe/amplitude modulation to encode a variable number of bits at 6 baud simultaneously on hundreds of carriers, each on a unique frequency, to achieve ~11 kbps (PEP1), ~18 kbps (PEP2), or ~23 kbps (TurboPEP). But those and other modulations were either half-duplex (PEP, V.29FT) or split the available frequencies either symmetrically (Bell 103J, Bell 212A, V.22bis) or asymmetrically (USRobotics' HST) in order to allow bidirectional communication.
Not being an expert in digital signal processing, I could not guess how hard it would be to accomplish but, with computing power practically exploding (pity all that power and more is wasted running bloated code, but that's off-topic) and the availability of computing clusters and grids, I wouldn't think that it would really be that hard for the modulations described above. (That said, I won't have it done by end of day ... or year ... or ...)
However, as others have pointed out, the real challenge with higher-speed modulations such as V.32 (9600 bps full-duplex) and V.34 (28800 & 33600 bps) is that, unlike the modulations listed above and in addition to the complicated modulations that you describe, they send and receive on the same frequencies and use echo cancellation in order to discriminate what the other side is sending from the echos of what the near side is sending. This adds another very complex layer and, without access to the transmitted data from at least one side, it might well prove to be impossible for a third party to decode what the other side is sending.
Geoffrey Welsh Never leave until tomorrow what can wait until next week.