Playground

Train a professional voice

Train a Professional voice from 10 minutes to 3 hours of your speech, confirm it is you with a spoken code, and get a notice and an email when it is ready.

The Professional tab on the Voice Cloning page trains a voice from a long recording of you, 10 minutes to 3 hours in total. It takes longer than an Instant clone and gives a closer match. You confirm it is you with a spoken code, then training runs while you do other work. You get a notice and an email when the voice is ready.

What you need

  • A workspace with credits. The page shows a price estimate before you confirm. The charges are on Voice Cloning limits and credits.
  • The clone.use permission. The default Owner, Admin and Member roles have it.
  • 10 minutes to 3 hours of your own speech, as files or as a recording. See Record a good sample for cloning.
  • A microphone the browser may use, for the code step.
  • No other professional voice in progress in the workspace.

Steps

Step 1: Open the Professional tab

Open Voice Cloning and choose the Professional tab. Pick the Dialect of your voice. The tab shows 10 min to 3 h, ready in about 30 min until you add audio.

Step 2: Add your audio

Upload or drag your files with Upload or drag files, or press Record and speak. Upload 10 minutes to 3 hours. Training needs at least 20 minutes of clear speech after cleanup, so aim for 30 minutes or more.

  • You can add up to 20 files, 2 GB in total.
  • The formats are WAV, FLAC, MP3, M4A, AAC, OGG, OPUS and WEBM.
  • A recording in the browser stops by itself at 30 minutes.
  • The bar under Total length shows what you have added, from 10 min to 3 h. It reads Add N more until you reach 10 minutes, and Over 3 hours if you add too much.
  • Each file shows Sent, Not sent or Cannot read this file. A file name can have up to 200 characters.

Step 3: Press Train voice

When the status reads Ready, check the price. The page shows This voice costs about N credits with 1 hour of training. The final price is set when training ends. Then press Train voice.

The page reads Starting, then Sending your files while the files upload. If some files do not send, the page reads Some files did not send. Retry them or go Back. Press Retry on each one.

Step 4: Confirm it is you

The page opens Confirm it is you with the same code step as an Instant clone. Press Record, say the sentence, then say each digit of the 4-digit code. The code recording is 3 to 10 seconds. The code expires after 10 minutes. Press New code for another one.

Step 5: Wait for training

Training starts. The page reads Professional voice and Training, with About N min left. You can leave this page; it will be in Your voices when it is ready. You can close the page. The voice has a Professional tag in Your voices. The first estimate, about 30 min, is a guess. It is not live progress, so the real time can differ.

Get it right

  • Speak naturally, in the dialect you picked, in one steady voice.
  • Use your own voice only.
  • Keep the room quiet. Noise and echo are copied too.
  • Send more than the minimum. 30 minutes or more leaves room for cleanup.

More advice is on Record a good sample for cloning.

Worked example

Omar uploads 3 files of his own speech, 45 minutes in all, with the Emirati dialect selected. The bar reads Ready, so he presses Train voice and reads the estimate. He says the sentence and the code, then closes the tab. A notice and an email tell him the voice is ready. His price:

  • Fee: $5.00, about 142,857 credits at the extra-credit rate.
  • Training: 1,800 credits per hour.
  • His audio: 1,100 credits per minute for transcription and cleaning, so 49,500.
  • The checks and the preview line are extra. The full line list is on Voice Cloning limits and credits.

What happens next

You get an in-app notice and an email when the voice is ready or when it fails. The link in the ready message opens Text to Speech with the voice selected.

  • Statuses. The status of a professional voice shows in Your voices.
Status What it means
Uploading Your files are being sent, or the upload was left unfinished
Training The voice is being trained. The page shows Training, N min left, an estimate
Failed Training did not finish. See Troubleshooting for the reason
  • Where the voice works. Text to Speech (the Cloned filter), the Voices page under Recorded by you, Assets on its Voice Cloning tab and EchoReader. See the Voice Library and Generate speech. A cloned voice does not work in agents yet.
  • Rename. You can rename the voice while it uploads or trains. A name has up to 60 characters, unique in the workspace. Near the end the page reads This voice is almost ready. Rename it in Your voices.
  • Remove a failed or unfinished upload. Open the voice in Your voices and press Delete. A failed voice also has Try again. During the upload, press Back and choose Cancel upload.
  • Delete a ready voice. Open the voice and press Delete. The voice is deleted right away. Anything already made with it stays where it is. The consent recording is kept 12 months. Deleting frees the slot and the name at once.

Limits

Limit Value
Total audio 10 min to 3 h
Clear speech after cleanup At least 20 minutes, or training fails
Files Up to 20
Size 2 GB in total
File name Up to 200 characters
Formats WAV, FLAC, MP3, M4A, AAC, OGG, OPUS, WEBM
Recording in the browser 30 minutes
In progress One professional voice per workspace
Voices per workspace 20. A professional voice in progress counts as one
Stopping a job A training job cannot be stopped
Failed job Refunded in full
Codes per user 20 per 10 minutes
Code checks per user 10 per 10 minutes

The Professional tab lists its own dialects. They are: Emirati, Saudi, Egyptian, Kuwaiti, Fusha, American, British and English. Qatari, Omani, Lebanese and Syrian are Professional only. Iraqi has no cloning.

Troubleshooting

A professional voice is already being made. Wait for it to finish.

Your workspace trains one professional voice at a time. Wait for it to be ready or to fail. If it is an unfinished upload, open it in Your voices and press Delete, or press Back and choose Cancel upload. Then press Train voice again.

Professional needs 10 minutes to 3 hours in total.

The files add up to less than 10 minutes or more than 3 hours. Add more audio, or remove some, until the bar reads Ready.

Professional cloning is not available right now.

The speech service has Professional switched off. Nothing was started. Try again later.

Less than 20 minutes of clear speech was left after cleanup. Add more audio.

The files had enough minutes, but too little clear speech. Train again with 30 minutes or more of clear speech. A failed job is refunded in full.

A workspace can keep up to 20 cloned voices. Delete one first.

The workspace is full. Delete a voice you no longer use, then press Train voice again. Deleting frees the slot at once.

Professional cloning is busy right now. Try again later.

The training service is full. Try again in a few minutes.

Training could not start. Try again later.

Try again later.

Training failed. Try again, or contact support with the job id.

Training did not finish. Try again, or contact support and give the job id.

Training was interrupted on our side. Try again.

Training was cut short. The page can also read "Training was interrupted. Try again." Try again. A failed job is refunded in full.

Training took too long and was stopped.

Training ran too long and was stopped. Try again.

The audio must be 10 min to 3 h of speech.

Keep the total between 10 minutes and 3 hours.

The files are larger than 2 GB.

Remove files until the total is 2 GB or less.

We could not read one of the audio files.

A file is damaged or in an unsupported format. Remove it or export it again.

The training audio is not the same person as your recording.

The code voice is not the voice in the files. Use your own voice for both.

We could not confirm it is you. Record again.

The code recording did not pass the check. Record the new code again in a quiet room.

That dialect is not supported.

Pick another dialect from the list.

That name was taken while training ran. Try again with another name.

Another voice took the name. Try again with another name.

Not enough credits to train this voice. Top up and try again.

The balance cannot pay for it. Top up, then train again.

Training finished but the voice could not be published. Try again.

The voice was not added to your voices. Try again.

The voice was trained but could not be tested. Try again.

The preview line failed. Try again.

Price not set yet. Voice cloning opens once its price is published.

Cloning has no price yet, so it is closed. Nothing was charged. Try again later.

FAQ

How long does training take?

The page says ready in about 30 min before you start, and About N min left while it runs. Both are estimates. You can leave the page. A notice and an email tell you when it is done.

Am I charged if training fails?

No. A failed job is refunded in full. The 1 hour estimate is charged first. When training ends, only the difference is charged or refunded.

Can I stop a training job?

No. You can remove a failed or unfinished upload with Delete in Your voices.

Can I train two voices at once?

No. One professional voice can be in progress per workspace.

How much does it cost?

A $5.00 fee (about 142,857 credits at the extra-credit rate, 157,895 at the Starter rate), plus 1,800 credits per training hour, plus 1,100 credits per minute of your audio for transcription and cleaning (3 hours is 198,000). The checks and the preview line are extra. The page shows an estimate first. See Voice Cloning limits and credits. Using a cloned voice in Text to Speech costs the same as any voice.