Use a library voice
Choose this for speed, low setup effort, broad testing, or when the agent does not need to sound like a particular authorized person.
Academy · Voices
Choose, clone, and test the right voice
Match a voice to the agent’s language and job, preview library voices, create a consented clone from a clean 20–30 second sample, assign it, tune delivery, and verify it in a real conversation.
A voice is a comprehension and trust choice, not decoration. Begin with the agent’s language, audience, speaking environment, and job. Preview a library voice first; clone a voice only when you have the speaker’s consent and a genuine reason to preserve that identity.
Consent comes before recording
Clone only a voice you own or have clear permission to use. Do not imitate another person, public figure, employee, customer, or spokesperson without authorization.
01
Choose a library voice or a clone
The best default is the simplest option that sounds clear in the required language and can be governed responsibly.
Create a cloned voice
Choose this when an authorized brand or speaker identity matters, you can record clean speech, and you are prepared to review consent, cost, and quality.
Language
Instead of
Choose by gender or name alone.
Use
Filter for the language, then test code-switching, names, numbers, and local place names.
Job
Instead of
Pick the most dramatic voice.
Use
Match pace and register to the job: calm support, clear reception, or energetic outreach.
Environment
Instead of
Judge from headphones only.
Use
Preview on a phone speaker and in moderate background noise.
Preference
Instead of
Approve from one sentence.
Use
Test a greeting, a question, an amount, a date, a correction, and a closing.
02
Preview a library voice properly
Open the voice picker in Assistant Details. Filter and preview before assigning; a matching language code is necessary but not sufficient.
Filter by primary language
Start with the language the agent will use most. If the agent is multilingual, test each configured language.
Shortlist two or three voices
Avoid reviewing dozens without criteria. Compare clarity, pace, warmth, and pronunciation against the same script.
Preview telephone content
Use names, amounts, addresses, dates, abbreviations, and a natural closing—not only a marketing sentence.
Assign and save
Choose the best-understood voice, not merely the most expressive voice, then verify that it remains selected after navigation.
Voice comparison script
Hello, this is Maya from Acme Properties.
I heard a budget of eighty lakh rupees and a preferred location near Gachibowli.
Did I understand that correctly?
Would tomorrow, Friday the fourteenth, between ten and eleven in the morning work?
Thank you. A property expert will follow up after your confirmation.03
Prepare a clean 20–30 second sample
HeyRik’s cloning flow recommends about 20–30 seconds of natural speech. Quality matters more than making the clip long.
One authorized speaker
Use one person from beginning to end. Confirm their consent and your right to deploy the resulting voice.
Natural, varied speech
Speak at a normal pace with complete sentences, several vowel sounds, and a short number sequence.
Quiet recording
Use a quiet room, stable microphone distance, and no music, echo, overlapping voices, or aggressive noise reduction.
Listen before upload
Reject clips with clipping, long silence, volume jumps, mouth noise, background speech, or a cut-off beginning or end.
English sample script — about 20–30 seconds
Hello, and thank you for helping create this voice. I am recording a short
sample so the assistant can learn how I speak. I usually talk in a calm and
friendly way, taking my time with each word. Today is a bright morning, and I
am reading these lines clearly and naturally. One, two, three, four, five.
Thank you very much.Use the script shown for the selected language
Read naturally rather than performing. If the speaker stumbles, record again instead of stitching multiple takes with different volume or microphone distance.
04
Create the clone in the dashboard
Open Voices, start the clone flow, and review every field and cost before submitting.
Choose Clone Voice
Give the voice a recognizable internal name that identifies its authorized owner and intended use.
Select language and profile details
Choose the primary language of the sample. Profile fields help filtering; they do not repair a mismatched recording.
Record or upload
Record in the dashboard or upload a clean supported audio file. Aim for the recommended 20–30 seconds.
Confirm consent
Use the consent control only after confirming the speaker understands and permits the intended cloning and deployment.
Check the displayed credit price
The current product is the authority for the charge. Review the price shown immediately before submitting; do not rely on a hard-coded docs amount.
Wait for Ready
Do not assign a training or failed voice. If training fails, correct the recording issue and retrain rather than repeatedly submitting the same clip.
Why the Academy does not state a fixed clone price
Existing sources disagree on a numeric credit charge. The dashboard’s displayed price at submission is the value the user must review and accept.
05
Assign the voice and tune delivery
Return to the agent’s Assistant Details to choose the clone. Tune speaking behavior in Call Settings, where it can be tested beside listening and interruption controls.
Voice
Pick the Ready clone under My Voices or choose a library voice.
Assistant DetailsLanguages
Make sure the agent’s primary language matches what the voice can speak clearly.
Assistant DetailsSpeaking speed
Begin near natural speed; change gradually and retest comprehension.
Call SettingsDelivery
Choose a restrained mood appropriate for the job and caller context.
Call SettingsPersonality
Control how replies are worded; do not use it to replace explicit prompt instructions.
Call SettingsInterruptions
Test whether callers can naturally interrupt without the agent losing the conversation.
Call SettingsAPI equivalent
List or filter voices with GET /voices, clone with POST /voices/clone, preview with POST /voices/preview, and assign the returned voice ID in the agent’s voice configuration.
06
Test the voice in conversation
A preview proves the voice can synthesize one line. A test call proves whether it works with turn-taking, your prompt, caller interruptions, and telephone audio.
07
Fix the source of a voice problem
Do not repeatedly adjust every control. Identify whether the defect comes from the recording, selected language, delivery setting, prompt wording, or turn-taking.
Muffled clone
Instead of
Increase speed or emotion.
Use
Record again in a quieter room with stable distance and clean input level.
Wrong pronunciation
Instead of
Assume a longer sample fixes everything.
Use
Test a voice suited to the language and write ambiguous names or abbreviations in a speakable form where appropriate.
Robotic pacing
Instead of
Rewrite the whole prompt.
Use
Adjust speed gradually and shorten long agent turns.
Talks over caller
Instead of
Change the voice.
Use
Tune wait-before-reply and interruption settings using natural pauses.
Clone unavailable
Instead of
Assign it while training.
Use
Wait for Ready; inspect the failure, correct the clip, and retrain.
Unexpected charge
Instead of
Rely on an old docs number.
Use
Stop before submission and review the currently displayed credit price.