Gemini 2.5 Pro TTS (Achird) regression: Audio bumps and pitch jumps between sentences

Hi team,
​I have been using the Gemini 2.5 Pro preview TTS model with the Achird voice for content narration. Previously, the output was consistent and clear.
​In recent builds, there is a noticeable regression:
​Sudden volume bumps and pitch jumping at sentence boundaries.
​Degraded voice clarity compared to earlier versions.
​Is this a known checkpoint issue in the preview model? Are there any recommended prosody/style parameters or stable checkpoints available to eliminate these sentence jumps?