Design Arena's Audio Realism Bench results of 4 August 2026 ranked Bland Speech v3 first of 13 AI voices at 1365, with the human recording at 1500, and each model in them was represented by one female and one male voice chosen as its most naturally conversational American-English voice. A clone of a particular person's voice was not what got scored, so the result does not tell a buyer how a clone will sound on a call.
That leaves a phone agent with a real choice, and most of it is about the work around a voice and not the voice itself: a stock voice arrives ready, while a clone arrives with a recording session, a consent step, and a person whose voice it is.
Should a phone agent use a cloned voice or a stock voice?
A phone agent should use a stock voice unless a specific person's voice is the point, because a stock voice needs no recording session and no consent step from the business, while a clone needs both.
Bland Speech gives every organization three built-in voices, Karen, Matthew, and River, and a public library of more, so a phone agent can be live without collecting a recording from anyone. A clone earns its extra work when callers should hear someone in particular: a founder, a named receptionist, or a brand's long-time voice actor.
Use a stock voice when
- A team wants a phone agent live without collecting recordings or consent from anyone
- A team will choose from a library of voices, some of which carry a per-character creator fee
Use a cloned voice when
- A team will gather one clean sample of about 10 seconds for a test, or 30 minutes or more of audio for a professional clone
- A team will keep a speaker's confirmation on record and decide in advance what happens if the speaker withdraws
How does each sound on a phone call?
A stock voice sounds the way its model was built and tested to sound, and a cloned voice sounds like its recording, because Mistral says a clone picks up the emotion and pace of its source and also the background noise and the microphone.
That is why the realism result describes the voices Design Arena scored and not a clone of anyone's voice. A sample recorded beside a fan brings the fan along. Cartesia says up to 60 seconds of audio better retains a speaker's accent, Speechify says a voice keeps the accent of the recording it was made from, and Descript says its clones come from a model based on US English pronunciation, so accents may not be preserved accurately.
Phone audio narrows everything. Bland Speech offers mu-law, a compressed encoding used on phone lines, at 8 kHz only and labels it for telephony, Grok TTS labels 8000 Hz as narrowband for telephony, and Inworld calls mu-law its compressed telephony format. Bland Speech v3 renders natively at 48 kHz, so a call receives a version resampled down to 8 kHz. A business that is choosing between two voices listens to both in the format callers will hear, and the difference between mu-law and PCM decides what that format is. Bland Speech delivers both through its text to speech API.
What does each option cost and require?
A stock voice at Bland Speech costs $0.04 per 1,000 characters, pay as you go with no monthly plan and 50,000 free characters once for a new account, though some public-library voices add a per-character creator fee, while a clone adds a recording, a confirmation, and for professional clones a review.
An instant clone at Bland Speech needs one clean sample of about 10 seconds and is ready in about a minute. A professional clone is fine-tuned on 30 minutes or more of audio and goes through Bland's consent and rights review, and every clone is private to the customer's organization. How much audio it takes sets those numbers beside other providers.
Rivals put plans in front of cloning. ElevenLabs starts instant cloning on its Starter plan and professional cloning on Creator, Cartesia starts professional cloning on Startup, and Murf offers instant cloning on its Enterprise plan only. Cartesia bills speech from a clone at 1 credit per character, and MiniMax charges $1.5 per cloned voice when the voice is first used. These are rivals' published terms as of November 2026.
What can go wrong with a clone that a stock voice avoids?
A clone can fail on consent and on portability, two things a stock voice never raises, because a clone is built from a real person's recording.
Consent comes first. Providers ask for a confirmation, a spoken challenge phrase, a signed agreement, or a recorded statement, and the consent rules by provider are laid out with the evidence for each. Descript says it cannot create voice clones for a deceased individual, and OpenAI's usage policies require a clear disclosure to end users that a TTS voice is AI-generated, so a cloned voice on a call is still an AI voice and a business plans for that.
Portability comes next. Typecast's API terms bar exporting a cloned voice outside its APIs, and a business that changes provider re-clones from the original recordings with the speaker's consent.
Which should a phone agent use, a cloned voice or a stock voice?
Bland Speech fits a phone agent on either: stock voices for an agent that needs no particular person's voice, and an instant or professional clone for an agent that should sound like someone a caller knows.
The sensible order is stock first and a clone second. A stock voice gets a phone agent live on Bland Speech v3, which Design Arena's listeners ranked first of 13 AI voices in results of 4 August 2026, and a clone follows once a business has a speaker, a signed confirmation, and a reason. Bland Speech is built for phone calls, and its clones stay private to the customer's organization.
The best voice cloning for phone calls ranks providers on exactly this use. The voice cloning comparison hub and the wider text to speech comparison set them side by side, and Bland Speech has a studio for trying a voice before building on it.
Frequently asked questions
Does a cloned voice sound as human as a stock voice?
A clone's realism follows its recording. Design Arena's results of 4 August 2026 scored one female and one male voice per model, each chosen as that model's most naturally conversational American-English voice, and a customer's clone was not part of that selection.
Which voices does Bland Speech give every organization?
Bland Speech gives every organization three voices, Karen, Matthew, and River, and its public library adds more. Bland's speech page also shows Valentine, Morgan, Tatum, Mackayla, and David.
Do public-library voices cost extra?
Some do. Bland charges a per-character creator fee on some public-library voices, on top of the standard rate of $0.04 per 1,000 characters.
Can a phone agent use a founder's voice?
A phone agent can use a founder's voice through a Bland Speech clone when the founder's recording is clean and the voice is confirmed as theirs to use. An instant clone needs about 10 seconds, a professional clone needs 30 minutes or more and a consent and rights review, and the clone is private to the customer's organization.
What audio format does a phone call use?
A phone call usually carries 8 kHz audio. Bland Speech's mu-law option is 8-bit and 8 kHz only, labelled for telephony, and its API also offers PCM at 8, 16, 24, 44.1, and 48 kHz.
Can another customer use a clone made with Bland Speech?
No. Bland says clones are private to the customer's organization, and the customer confirms every voice is theirs to use.
