VowelMarks

Persian dialogue

Two Persian voices. One conversation.

Two friends notice a bud on a branch they thought was dead. Give one the observation and the other the surprised reply, then listen to whether the exchange works without the speaker labels.

Researched and edited by VowelMarks · Last reviewed

Start with turns, not character labels inside the text

Keep each person's Persian line in its own speaker turn. Choose a voice for each role, direct their delivery separately, then listen to the whole exchange. VowelMarks keeps the speaker assignments separate from the Persian script, so names such as Kimiya and Aava identify the turns without becoming spoken words.

A tiny scene worth testing

One person spots a bud on a dry-looking branch. Their friend had given up on the tree. Three lines are enough to hear whether those two reactions come through. This is an original test scene, not a quotation from a story or a claim about a finished recording.

Kimiya

ببین، روی این شاخه یک جوانه هست.

Aava

فکر می‌کردم این درخت خشک شده است!

Kimiya

هنوز زنده است. بهار از همین‌جا شروع می‌شود.

The exchange is: “Look, there is a bud on this branch.” “I thought this tree had dried up!” “It is still alive. Spring starts right here.” The names above are editor labels, not lines for the voice to read. You do not have to add Kimiya: to the Persian sentence.

Open this script with both voices

Give the roles different jobs

Kimiya is noticing something small, not delivering a speech. Aava is reacting to it, not narrating the entire scene. Start with restrained direction for the first role and a more conversational delivery for the second. A different voice helps the listener keep track; different direction gives the voices a reason to sound different.

Kimiya: Director’s notes

A calm Iranian Persian narrator noticing a small sign of spring. Warm, close and restrained.

Aava: Director’s notes

An Iranian Persian garden visitor, curious and pleasantly surprised. Speak naturally, without shouting.

The starter uses Kimiya at a slightly measured speaking pace and Aava at a natural pace. Those settings direct the performance you are generating. The playback control is separate: it lets you listen to the finished exchange faster or slower without generating it again.

Build the exchange in the Speech workspace

  1. Choose Premium and open the voice picker for the first turn. Select the first person's voice.
  2. Write that person's Persian line. Use Add speaker below the editor for the reply.
  3. Choose the second voice and write the reply. Use Add passage for the next turn, then select the first person again.
  4. Open the selected speaker's Studio settings. Adjust delivery, speaking pace and Director’s notes for that role.
  5. Generate the short exchange. Listen to the transitions before expanding the conversation.

Multiple turns do not mean multiple new voices. This workflow is a conversation between two speakers, and either speaker can return later in the script. The role assignment is kept apart from the source words, including right-to-left Persian and paragraph spacing.

A cue belongs to a moment, not the whole conversation

The starter puts Curious on Aava's reply. It does not ask Kimiya to become curious too. Select the Persian words you want to change, then insert the cue. A pause or sigh belongs at a cursor position; a delivery cue belongs to a phrase.

A global note such as “quiet and restrained” describes the usual performance. A local cue asks for an exception on the marked words. Keep the exception intelligible: a surprised question fits this scene better than laughter on every line. The cue directs the model; listen to the result to judge whether the change is actually audible.

Check the recording as a conversation

  • Who is speaking? Hide the labels and listen. Can you still follow the exchange?
  • Where is the handoff? The reply should arrive after the first line, without swallowing its ending.
  • Are the Persian readings right? Check the words in context, especially names and Ezafe, before judging emotion.
  • Does the cue fit? Compare Aava's surprised reply with Kimiya's quieter return.
  • Is the export complete? Play the downloaded file from its opening to its last word.

For a listening lesson, ask “Why is the second person surprised?” before showing the text. For a story, keep the script, voice choices and direction with the reviewed take so the next scene starts from the same choices.

Keep the web workflow and API contract separate

A provider's two-speaker capability is not automatically a field you can send to VowelMarks' public API. Developers should start with the documentedspeech request contract andPython and JavaScript examples. This guide describes the Speech Studio conversation editor, not an undocumented API request.

Where to go next

Frequently asked questions

How do I make a Persian dialogue with two AI voices?

In the Speech Studio dialogue workflow, add a speaker turn, assign a voice to each person, and write each person's Persian lines in their turn. Set the delivery and direction for each speaker, then listen to the complete exchange before exporting.

Is a speaker the same as a voice?

A voice is the underlying vocal sound. A speaker is one role in the conversation, with a selected voice, delivery, speaking pace and Director’s notes. Speaker names identify turns; they are not part of the spoken Persian script.

Can I use the same voice for both speakers?

Yes, but the listener may find two different voices easier to follow. Direction can distinguish roles, while the voice choice makes the basic vocal difference. Listen for a clear handoff rather than relying only on the labels in the editor.

Can I make a Persian conversation in the Developer API?

The public Developer API speech documentation defines its supported request fields. The Speech Studio dialogue workflow described here does not by itself add a public multi-speaker API contract.

Sources and review notes

Sources are listed for the claims they support. Original practice examples are identified in the article. Product capabilities and access terms were checked on the review date and can change.

  1. Google AI for Developers, Speech generation. Provider documentation for Persian speech, two-speaker generation and performance direction. Provider capabilities are separate from features released in VowelMarks.
  2. VowelMarks, Direct an AI Persian voice. The distinction between a voice, whole-passage direction and cues attached to particular Persian words.