## v1.3.014 — 2026-07-14
- fix: **Audio from the "Remove background music" option when creating a cloned voice no longer contains noticeable distortion or noise.**
- fix: **Custom and cloned voices no longer fall silent or revert to a built-in voice in certain cases.** This covers migrating voice settings after an update, recognizing cloud cloned voices, and automatically recovering when a previously selected voice becomes unavailable.
- fix: **Turning off "deep thinking" for DeepSeek models now takes effect correctly.**
- improvement: **When a cloud model returns no usable content, the chat now shows a clear message that distinguishes the likely cause** — for example, the reasoning process using up the reply-length limit, or reaching the length limit — instead of leaving an unresponsive, blank turn.
- improvement: **A sentence that fails during local voice synthesis now shows a brief notice,** making it clear the sentence was skipped rather than passing silently.
- improvement: **Reference-audio handling for voice cloning and overall voice-output quality have been refined,** reducing distortion introduced during volume adjustment.