So in that sense it is quite different from the account or UTXO logic that most cryptocurrencies use, where balances can be transferred between accounts indefinitely. I think it's interesting to think about what problems it solves - it semi-solves anonymity in a way that might be acceptable for lawmakers and that doesn't require a backdoor (the more typical approach for 'compliant privacy'). The question is whether that is enough utility gain for regular users to justify the overhead - it seems like a simple transaction requires quite a lot of steps, key management is a non-trivial burden (for users who might not even be used to password managers eg),...
The schema they use to explain the system's mechanisms clearly shows an exchange that needs to be audited:
> Payments are made after exchanging existing money into electronic money with the help of an Exchange service, that is, a payment service provider for Taler.
That is, the transaction can be offline, but it eventually needs to pass through a central point to be effective.
This.
On the contrary, GNU Taler is only a transaction system as you said, and uses advanced cryptography (double blind signatures) and infrastructure that allow it to be bearer-based, i.e., actually decentralized transactions between two offline parties can take place. Taler can actually do digital cash, with buyers privacy, and auditability and taxability of sellers.