AI & TechArtificial IntelligenceBigTech CompaniesNewswireTechnology

Gemini Live’s AI voices sound unnatural

▼ Summary

– Google’s AI model updates are causing disruptive changes to the Gemini Live assistant’s customizable voice options.
– Voice characteristics like cadence, tone, and preset accents are reported to change frequently, sometimes week to week.
– Specific voice options, such as the “Capella” British accent, have significantly deteriorated or broken down during use.
– The actual spoken voice often mismatches its preview, with slower patterns, altered pitch, and unintended accent shifts during conversation.
– Audio artifacts like crackles and pops are a growing complaint, though the issue is sporadic and not present in all contexts like Android Auto.

The recent rollout of Google’s latest AI models is having a significant, and often frustrating, impact on the user experience of Gemini Live and its customizable voice options. While the Gemini Live 3.1 Flash Live update might seem like the obvious culprit, this pattern of vocal instability is not new. Users frequently report that the assistant’s preset voices undergo noticeable shifts in cadence, tone, and even their designated accents from one week to the next.

One of the most pronounced changes involves the overall speech rhythm. In practical testing, the Capella” voice, designed to emulate a female British accent, has degraded considerably since its debut. This issue is not isolated, as numerous other regional voice presets within Gemini Live demonstrate similar problems. Over recent months, the quality of many options has declined, with the brief preview audio sounding markedly different from the actual conversational experience.

During live interactions, speech patterns often become unnaturally slow, and distinctive vocal characteristics are muted. A high-pitched voice might be toned down, or an accent may unpredictably drift, sometimes sounding Australian or defaulting to a more generic American accent mid-conversation. This creates a disjointed and inconsistent interaction.

A temporary fix involves resetting the Gemini application, which can briefly restore the selected accent. However, the voice typically begins to slowly morph again during use, resulting in an awkward hybrid sound that users find jarring. Compounding these issues are sporadic audio artifacts like crackles, pops, and hisses, which have become a growing point of discussion on Google’s support forums, though they are not consistently reproducible across all voice options.

Interestingly, the problem appears context-dependent. Many voice options perform normally when used for simple voice commands or when accessing the Live feature through Android Auto in a vehicle. It remains unclear whether Google’s development team is actively aware of these specific performance degradations. Inquiries have been made to the company for clarification, and any substantive response will provide further insight into the situation.

(Source: 9to5google.com)

Topics

gemini live updates 95% voice option changes 93% speech cadence issues 90% accent deterioration 88% audio artefacts 85% user complaints 82% gemini application reset 80% voice preview discrepancy 78% android auto integration 75% google ai models 73%