Custom Voice Cloning (Beta)

When a knowledge seeker plays one of your agent's answers, it is read aloud in a voice you choose from the stock voices in your advanced agent settings. With custom voice cloning, your agent can read its answers in your own voice instead. Your audience hears you, not a narrator.

Beta

Custom voice cloning is a beta feature. The exact form it takes will evolve as we get feedback, and for now the request is handled by our support team rather than from within the app.

What a custom voice can do

A custom voice is built from a short recording of you speaking, about ten seconds long. Once it is attached to your agent:

• Every answer your agent reads aloud uses your voice, streaming as it is generated, so listening still starts within a few seconds.
• Long answers are read in full, with natural pacing and pauses rather than a flat recitation.
• The voice is built for the language you record in, and can carry over to a small number of other languages when your agent answers in them.

Limitations to know about

• The clone is only as good as the recording. Background noise, a distant microphone, or a rushed read all come through in the result.
• One primary language. Your voice is built for the language you record in. Carrying it into other languages is limited to a handful of pairings, and we will let you know if a language you need is not yet supported.
• A consent recording is required. You must record yourself reading a fixed consent statement, word for word. We cannot build a voice without it.
• Playback only. Your voice is used solely to read your agent's answers to your Knowledge Seekers.
• Turnaround is manual. Allow up to five business days for us to attach your custom voice to your agent.

Before you begin

You will record two short clips. A mobile phone records perfectly good audio for this, so there is no need for special equipment. For the best result:

• Find a quiet room with soft surroundings. Avoid echoey spaces, fans, traffic, and other people talking.
• Use the same device, in the same place, for both clips.
• Keep each clip to ten seconds or less.
• Speak alone. No music, no other voices.
• Standard phone recordings in m4a, mp3, or wav format are all fine.

Step 1. Record the consent clip

Custom voice cloning is powered by Google, which requires a spoken statement that you own the voice being cloned. Record yourself reading the following, exactly as written:

I am the owner of this voice and I consent to Google using this voice to create a synthetic voice model.

The wording cannot be changed or shortened. If your primary language is not US English, the statement has its own wording in your language. Tell us your language when you send your request and we will reply with the text to record.

Step 2. Record the reference clip

The reference clip is what your voice is built from, so this is the one to spend a moment on. Record up to ten seconds of yourself speaking naturally, in the tone you would like your agent to have. A good choice is to introduce yourself and describe what you help people with, the way you would to someone you had just met.

• Be a little more energetic than you want the result to sound. The clone tends to come out slightly calmer than the recording.
• Use natural pauses and pacing. Speak as you would in conversation, not as if reading a list.
• Avoid a monotone. Some variation in pitch and emphasis gives the clone something to work with.
• Stay close to the microphone and record in the same room as the consent clip.

Step 3. Send your request

1. Attach both clips to an email to support@answersfrom.me
2. Put Custom Voice Cloning Request in the subject line
3. If your primary language is anything other than US English, say so in the body of the email

If you recorded on your phone, the easiest route is to open each clip in the phone's voice recorder app, tap Share, and choose your email app. That attaches the file to a new message, and you can add the second clip the same way before sending.

What happens next

Allow up to five business days. We will build your custom voice, attach it to your agent, and let you know when it is ready. To hear it, open any conversation with your agent and play an answer.

If you are not happy with how it sounds, reply to the same email thread. You can send a new reference clip to try again, or ask us to switch your agent back to one of the stock voices.