I interpret this question as asking, what is the first ordinal $\alpha$ such that there is some lost melody (in the sense of Hamkins-Lewis Theorem 4.9) in $L_\alpha$.

The answer is $\alpha=\Sigma+1$ (or $\Sigma+\omega$ if you really insist on it being limit). Indeed, the argument in Theorem 4.9 of the paper produces a lost melody inside $L_{\beta+1}$, where $\beta$ is the least ordinal with the following property: $\beta$ is countable inside $L_{\beta+1}$, and every ITTM either halts or repeats by time $\beta$. From Welch's work we can deduce $\beta=\Sigma$: indeed, by the argument in Corollary 2.3, every machine repeats by time $\Sigma$, and to see $\Sigma$ is countable in $L_{\Sigma+1}$ we can use the fact that $\Sigma$ is not admissible (Corollary 3.4) to get a definable sequence $\alpha_n$ in $L_\Sigma$ tending to $\Sigma$. The elements $\alpha_n$ are countable (and uniformly so) because we can pick out a bijection to $\mathbb N$ defined by their first occurrence of a real coding $\alpha_n$ in some machine listing all accidentally writeable reals. Therefore $\beta=\Sigma$ and so we have a lost melody in $L_{\Sigma+1}$.

On the other hand, all reals in $L_\Sigma$ are accidentally writeable, since they are computable from accidentally writeable ordinals (as explained in the proof of Theorem 4.9) and accidentally writeable reals can't be lost melodies (since we can list all accidentally writeable reals and check them one by one).