All 30 Gemini voices saying the same sentence in Spanish. Hear them in studio quality and, with one click, as they would sound on a call.
Original quality, 24 kHz: this is how they sound in the browser.
The same take through a phone line (8 kHz): this is how the caller hears it.
No voices match that combination of filters.
Every sample says, in Spanish: "Hi, I'm Phonica's voice assistant. I answer your calls, book appointments and resolve questions at any hour. How can I help you?". Generated with the gemini-3.8-flash-tts model. The name, one-word character and gender of each voice are as stated by Google; the grouping by character is ours.
A call travels at 8 kHz, compared with the 24 kHz of the samples. The highs are lost and the voice sounds thinner. The "Over the phone" switch reproduces exactly that cut: same take, different channel.
These samples are read-aloud recordings. On a real call the agent speaks in real time, reacting to the caller, and the intonation changes. The timbre of the voice, however, stays the same.
The voice only provides the timbre. Pace, warmth and manner of speaking are defined when the agent is configured: two agents with the same voice can sound very different.
For answering and handling calls, warm and soft voices tend to inspire more trust. But the one that fits you wins: filter, listen and compare.
What sounds full in the browser can change on a call. Deeper, warmer voices hold up better to the cut of the line.
Pick two or three candidates and try them in your own agent, with your script and a real call. It is the only comparison that truly counts.