While self-talk represents a promising Mechanism of Action (MoA) for anxiety management, the unfamiliarity of hearing one's own voice poses significant challenges for therapeutic engagement. Through two empirical studies (N=60, N=119), we examined a novel digital therapeutic approach combining Acceptance and Commitment Therapy (ACT) with self-talk externalization. Our first study confirmed self-talk's therapeutic potential but highlighted user discomfort with voice experiences. To address this, we developed a systematic voice equalization algorithm using Analytical Hierarchy Process (AHP) to identify optimal voice modifications through progressive refinement of 14 voice types. Our second study implementing this algorithm demonstrated both improved anxiety outcomes and enhanced user engagement. Participants reported that interacting with their modified voice facilitated deeper self-acceptance and emotional understanding. These findings suggest that self-talk, when properly integrated with voice equalization techniques, can effectively support anxiety reduction while promoting sustained therapeutic engagement in digital mental health interventions.