Book a demoDemoTalk to a specialist

Dashboard manual · Intelligence

Voice & tone: WhatsApp audio transcription and AI voice for audio replies

"Voz & tom" (Voice & tone) is the panel item that shows the audio received from leads with transcript and speech register, and where you choose the synthetic voice the AI uses to reply with audio on WhatsApp.

Who can use it

RoleWhat you see and do
OwnerSees and does everything: listens to the audio, changes the AI's voice, listens to previews, tests phrases and turns the voice reply on or off.
AdminListens to the audio, reads transcripts, listens to voice previews and tests phrases. Changing the voice and turning on the voice reply return an error notice: those are Owner-only actions.
AgentSees the menu item, the received audio, the transcript and the speech register. The voice previews and the phrase test do not play (they are for Admin and above) and the agent cannot change the voice.

Screens

"Voz & tom" (Voice & tone)

The "Voz & tom" (Voice & tone) screen is for following the path of a lead's audio (received, transcribed and analyzed) and defining how the AI's voice sounds when it replies with audio.

Address in the dashboard: /app/voz

The "Voz & tom" (Voice & tone) screen in the demo account, with sample data and blurred values.

1 · RECEIVED AUDIO

The audio that leads sent on WhatsApp, five per page, from newest to oldest.

Buttons and actions

"‹nome do lead›" (‹lead name›)
Clicking the audio card (or Enter/Space) selects it and shows its transcript and speech register alongside.
"Página anterior" (Previous page)
Goes back to the previous five audios.
"Próxima página" (Next page)
Moves on to the next five audios.
  • Each card: avatar, lead name, date and time, and the audio player (or Áudio indisponível (Audio unavailable)).
  • The selected audio has the brand-colored border.
  • Footer Page ‹n› of ‹total› · ‹n› audios when there is more than one page.

2 · TRANSCRIPT

The text of the selected audio, transcribed automatically.

  • Badge STT automático (Automatic STT).
  • The lead's speech in quotation marks, in italics.
  • With no text: Sem transcrição (No transcript) — the audio arrived before the transcript or it failed.

3 · SPEECH REGISTER

The lead's way of speaking, extracted from each transcribed audio.

  • A badge with the register (for example, informal) and tags with the speech markers detected.
  • A bubble with the suggested tone for replying to that lead.
  • With no analysis: Sem registro de fala ainda — é extraído automaticamente de cada áudio transcrito. (No speech register yet — it is extracted automatically from each transcribed audio.)

AI voice

Gallery of catalog synthetic voices, in three groups: "Femininas" (Female), "Masculinas" (Male) and "Neutra" (Neutral).

Buttons and actions

"Voz ‹nome›, ‹gênero›, ‹timbre›" (Voice ‹name›, ‹gender›, ‹timbre›)Owner
Clicking the voice row (or Enter/Space) makes the active persona use it; the Voz atual (Current voice) badge and the Voz da IA atualizada (AI voice updated) notice appear.
"Ouvir prévia da voz ‹nome›" (Play preview of voice ‹name›)Owner, Admin
Round play button: synthesizes a welcome phrase with the AI's name and plays it. Clicking again pauses (Pause voice preview ‹name›).
  • "Femininas" (Female): Nova (clear and energetic), Shimmer (soft and welcoming), Coral (warm and approachable) and Sage (serene and confident).
  • "Masculinas" (Male): Onyx (deep and full-bodied), Echo (firm and direct), Ash (young and friendly) and Ballad (calm and narrative).
  • "Neutra" (Neutral): Alloy (balanced and versatile), the voice used when none has been chosen.
  • Each voice has a colored circle that pulses while the preview plays (still when reduced motion is on).
  • Footnote: catalog synthetic voices; the lead's voice is never cloned or imitated (LGPD).

Test with your phrase

Lets you hear any phrase in the current voice, the same way it goes out on WhatsApp.

Fields and filters

"Frase de teste da voz" (Voice test phrase)
Up to 240 characters, with a ‹n›/240 counter. When empty, it plays the welcome phrase with the AI's name.
  • Player that synthesizes the phrase in the current voice; changing the voice or the phrase prepares a new audio.

"Responder em voz (TTS)" (Reply by voice (TTS))

Decides whether the AI replies with audio when the lead sends audio.

Buttons and actions

"Responder em voz (TTS)" (Reply by voice (TTS))Owner
A switch that takes effect immediately: when on, the reply to an audio goes out as audio, in the voice chosen above. It is disabled when there is no active persona.

Screen states

Loading
Gray blocks in place of the cards.
Nothing here yet
Nenhum áudio recebido ainda — os áudios dos leads no WhatsApp aparecem aqui. (No audio received yet — leads' WhatsApp audio appears here.)
If something goes wrong
Não foi possível carregar as amostras de voz (Could not load the voice samples), warning that the replies remain saved, with a button to try again.
Preview did not load
Notice Não foi possível carregar a prévia da voz. (Could not load the voice preview.) This is what the agent sees when trying to listen.
The browser blocked automatic sound
Notice Prévia pronta — toque de novo para ouvir. (Preview ready — tap again to listen.)
Voice change refused
Notice Não foi possível trocar a voz. Tente novamente. (Could not change the voice. Try again.)
No active persona
Nenhuma persona ativa encontrada. (No active persona found.) on the voice reply card, with the switch disabled.

Step by step

How to choose the AI's voice

  1. Open "Voz & tom" (Voice & tone) in the menu.
  2. On the "Voz da IA" (AI voice) card, click the play button of at least three voices to hear the preview.
  3. Click the row of the chosen voice; it gets the Voz atual (Current voice) badge.

How to hear a phrase in the AI's voice before turning it on

  1. On the "Teste com a sua frase" (Test with your phrase) card, type the phrase in the box.
  2. Press play on the player just below.

How to make the AI reply to audio with audio

  1. Choose the voice on the "Voz da IA" (AI voice) card.
  2. Turn on the "Responder em voz (TTS)" (Reply by voice (TTS)) switch.
  3. From then on, when the lead sends audio, the reply goes out as audio.

How to read the transcript of a lead's audio

  1. In "1 · ÁUDIOS RECEBIDOS" (1 · RECEIVED AUDIO), click the audio card.
  2. Read the text in "2 · TRANSCRIÇÃO" (2 · TRANSCRIPTION) and the way of speaking in "3 · REGISTRO DE FALA" (3 · SPEECH LOG).
  3. Use "Próxima página" (Next page) to see older audio.

Frequently asked questions

Does the AI reply with audio to every message?

No. With "Responder em voz (TTS)" (Reply by voice (TTS)) on, it replies with audio when the lead sends audio; text messages still get text.

How many voices are there to choose from?

Nine: four female, four male and one neutral.

Which voice does the AI use if I never chose one?

The neutral one, Alloy.

Can the AI speak in my voice or my salesperson's voice?

Not on this screen: the voices are catalog synthetic voices. For the lead to hear your team's voice, record replies in "Intent & catálogo" (Intents & catalog), on the "Respostas gravadas" (Recorded replies) card.

What is the maximum length of the test phrase?

240 characters; the counter beside the player shows how many have been used.

Why does an audio show "Sem transcrição" (No transcript)?

It arrived before the transcript was available or the transcription failed. The audio remains in the player for listening.

Does the speech register change the AI's reply?

It shows the lead's way of speaking and a suggested tone. The persona's tone, which applies to all conversations, is written in the "Editor de persona" (Persona editor).

Help center guide

"Voz & tom" (Voice & tone)

The audio messages leads send, what was said in them, and the voice your AI replies with. Two controls change behavior; the rest is read-only.

Many people reply by audio on WhatsApp, and this screen exists because of that. It shows what arrived as speech, what was said, and which voice your AI replies with. Of everything shown here, only two things change its behavior: the chosen voice and the "Responder em voz (TTS)" (Reply by voice (TTS)) switch — replying in audio, in everyday Portuguese. The rest is read-only.

The screen, part by part

On screen: "Voz & tom" (Voice & tone) screen with the received audio card on the left and the AI voice gallery on the right. No audio has arrived on this account yet: that is why the left side is empty and the transcription and speech log cards don't appear.

  1. The audio messages that arrived: The audio messages leads sent, from newest to oldest. Empty as here means nobody has sent audio yet — and the cards "2 · TRANSCRIÇÃO" (2 · TRANSCRIPTION) and "3 · REGISTRO DE FALA" (3 · SPEECH LOG) only appear after you select an audio from this list.
  2. The AI's voice: The first of the two real controls on the screen. The name shown in parentheses is your persona's: choose a voice that matches it.
  3. Nine voices in three groups: FEMININE (Nova, Shimmer, Coral, Sage), MASCULINE (Onyx, Echo, Ash, Ballad) and NEUTRAL (Alloy), each with its timbre described in two words. Click and it switches: there is no save button.
  4. The preview of each voice: Listen before you choose, as many times as you like. A preview does not deduct anything from your balance.
  5. The voice in use: The chosen voice has its card highlighted and the "Voz atual" (Current voice) badge. This is the voice sent on WhatsApp when replying by audio is turned on.
  6. The lead's voice is never copied: All voices come from a catalog. The audio of the person talking to you is used only to be understood — nothing is cloned or imitated.

1, 2 and 3 are the path of the audio

The three numbers are not tasks for you to carry out in order. They are the route an audio takes on its own when it arrives, shown step by step so you can check where it stopped:

1 · RECEIVED AUDIO
What leads sent by voice. Click one to see the next two cards for that audio.
2 · TRANSCRIPT
The same audio turned into text — this is how the AI understands what was said. "Sem transcrição" (No transcript) means that audio could not be read; the conversation goes on as normal, but the AI did not hear that part.
3 · SPEECH REGISTER
A reading of that person's way of speaking — formal, informal or neutral — with the expressions that appeared. It works as a mirror: if everyone speaks informally and your AI replies in a suit and tie, the mismatch is right here in plain sight.

Choosing the voice

  1. Listen to the previews: The round button next to each voice plays a sample. Listen to at least three before deciding — on paper the names tell you nothing.
  2. Test with your phrase: In the "Teste com a sua frase" (Test with your phrase) card, write up to 240 characters and listen in the current voice. Use a sentence your AI would really say, with the company name and a price in it: proper names and numbers are where voices differ the most.
  3. Click the voice: The choice is saved on click and applies from the next reply onward. Changing it does not affect any audio that has already been sent.
  4. Decide whether it replies by speaking: The "Responder em voz (TTS)" (Reply by voice (TTS)) switch, at the bottom of the column, makes the reply go out as audio when the lead sends audio. When it is off, the AI understands the audio and replies in writing.

With an A/B test running

The way of speaking is not adjusted here

Voice is the timbre; tone is the manner. This screen takes care of the timbre. There is no field, here or anywhere else, to choose between "you" and "sir", nor a control for reply length — if someone promised you that, the promise was wrong.

The way of speaking is controlled by writing, in the "Tom de voz" (Tone of voice) field of the persona, in "Configurações" (Settings) or in the Personas A/B editor. It is free text, and it is where instructions such as "address them informally", "replies of at most three lines" or "never promise a deadline without confirming" fit. The "Nome da IA" (AI name) also lives there, not here.

What costs money and what does not

  • Listening to a preview deducts nothing. You can listen to all nine, as many times as you like.
  • Testing with your own phrase deducts nothing. Write and listen freely before deciding.
  • Receiving a lead's audio is charged when it is turned into text. Converting speech to text is billed per minute and appears in the statement as Minuto de voz. A three-minute audio costs three times as much as a one-minute one. When the AI is handling the conversation, every audio is transcribed; when the team is handling it, only with "Transcrever áudios no atendimento humano" (Transcribe audio in human service) turned on in "Configurações" (Settings) — when it is off (the default), the agent listens and nothing is charged.

Your balance and the price of each item are in "Saldo & Recargas" (Balance & top-ups). It is worth checking after the first week with audio: an audience that talks a lot changes the month's bill.

Who can change things

  • Changing the voice and turning on reply-by-audio: only the account Owner.
  • Listening to previews and testing phrases: Admin and above.
  • Reading the audio, the transcript and the speech log: anyone with access to the screen.

In the dashboard: "Abrir Voz & tom" (Open Voice & tone)