Create Language-Learning Audio With Multiple Speakers

A vocabulary list can tell a learner what billete means. A two-person station scene lets them hear what happens when someone actually needs one.

That is the useful shape of language-learning audio: a small social situation, two distinct speakers, one practical goal, and enough space for the learner to notice how the exchange moves. Vois gives a teacher or course producer a local workspace to turn that reviewed dialogue into a repeatable clip with a stable cast. The voice generation comes after the teaching choice, not before it.

Start with a communicative scene, not a word list

A language-learning dialogue works best when each speaker needs something from the other person. The learner then has a reason to track questions, replies, confirmations, and changes of plan. That makes the audio easier to turn into a listening task than a string of unrelated example sentences.

Write a short brief before you draft the dialogue:

Decision	Example
Target language and learner range	Spanish, early intermediate
Scene	Buying a train ticket
Speaker goal	One person asks for a return ticket; the other confirms time and platform
Listening target	Recognize the destination, departure time, and platform
Follow-up task	Choose the right departure details, then answer as the station agent

The ACTFL World-Readiness Standards place communication and cultural competence at the center of language learning. The CEFR also frames learners as social agents. A compact scene is one practical way to give a clip that social purpose without turning it into an audio drama.

Review the target-language script before you generate it

Treat the target-language text as source material. Voice generation can render the words you provide, but it does not translate, proofread, choose an appropriate register, or verify cultural details for you.

Here is a Spanish station dialogue structure. It is a draft model, not a substitute for review by someone qualified to review the target language and learner level.

[Pasajera]: Buenos días. Quisiera un billete a Toledo para hoy, por favor.

[Empleado]: Claro. ¿Quiere salir por la mañana o por la tarde?

[Pasajera]: Por la mañana. ¿A qué hora sale el próximo tren?

[Empleado]: Sale a las diez y cuarto, desde el andén cuatro.

[Pasajera]: Perfecto. Un billete de ida y vuelta, por favor.

Ask the reviewer to check the parts that shape a learner's model of the language:

• Does the scene use the right level of formality for the location and relationship?
• Are the response choices natural for the situation, not merely grammatical?
• Are names, places, numbers, and abbreviations written the way the selected voice should say them?
• Does the clip contain only the language you want learners to hear and reuse?

Keep the speaker label only when the speaker changes. In a Vois dialogue script, [Pasajera]: and [Empleado]: give the cast a clear contract. Adding a label before every sentence from the same person creates unnecessary chunks and can make a natural turn sound chopped up.

Choose the language option deliberately

Choose the Vois language option that supports the reviewed target language, then make a short representative clip before producing a series. The multilingual option covers a focused language set; Pro offers the broader Omni language range. Either choice renders the text you supply. It does not translate, proofread, or validate the lesson.

Prompt your agent: "In Vois, inspect the available language options for this reviewed Spanish station dialogue. Set up a short test project using the option that supports Spanish. Keep the script unchanged, do not generate a full lesson set, and report the selected option, available voice choices, and any language-coverage limitation."

Review the result in this order:

1. Confirm that the selected option supports the target language and your current plan.
2. Have a qualified language reviewer listen to the names, numbers, rhythm, and register in the short test.
3. Record the approved option and any pronunciation decisions in the lesson's cast sheet.
4. Approve full production only after the test sounds appropriate for the learner level.

Language coverage is not a promise that an unreviewed script is ready to teach from. The test clip gives the reviewer the actual audio needed to make that decision.

Make the cast repeatable

For a recurring lesson series, a voice is part of the lesson's continuity. Learners should not have to wonder whether the station agent, shopkeeper, or host has been recast every time they open a new clip.

Start in the Vois voice library and save the exact voice ID for each approved role in the cast sheet. If an agent is helping with setup, ask it to show the available voices, match candidates to the two role descriptions, and preserve the exact IDs it selected. Do not let it silently substitute a new voice when a lesson is revised.

Role	Voice ID	Direction for the writer
Pasajera	narrators_ai_voices/morgan_vale	Asks for information, then confirms the choice
Empleado	hosts_ai_voices/alex_bright	Gives clear options and repeats the practical detail

The sheet is not a performance prescription. It is a production record. Add the target language, engine choice, and review date beside each role. When a listener reports that a particular name needs attention, you will know which script and cast decision to revisit.

You do not need voice cloning for this workflow. The library has 100+ voices, which is enough to begin with a clear, repeatable pair. If a project truly needs an authorized cloned voice, use a clear 10 to 15 second sample and get the voice owner's permission before you create it.

Build one dialogue project with an agent

Create one Vois project for the reviewed scene, then give an agent a constrained production request instead of a sequence of terminal commands.

Prompt your agent: "Create a Vois project named 'Spanish station dialogue.' Add the approved station-dialogue.txt script without changing its Spanish text. Create the two speakers, Pasajera and Empleado, with the exact approved voice IDs from the cast sheet. Generate only a review clip first. List the project, script, speaker, and export names you propose, and wait for approval before generating the complete dialogue or exporting it."

Use this human review and approval flow:

1. Compare the imported script with the language-reviewer's approved source, including speaker changes and paragraph breaks.
2. Check that each proposed speaker uses the exact voice ID in the cast sheet.
3. Listen to the review clip with the language reviewer, then correct the script or voice assignment if needed.
4. Approve the complete generation only after the reviewer signs off on pronunciation, pacing, and role distinction.
5. Review the final clips and transcript together, then approve the export for the learning platform.

This keeps the multi-speaker workflow useful for learner-facing dialogue without letting automation bypass the teaching and language decisions.

Use pronunciation review before batch production

A learner will copy what they hear. That is a reason to review proper names, place names, numbers, acronyms, and unfamiliar words before you make ten variations of the same scene.

Keep a small review list with four columns: written form, intended spoken form, surrounding sentence, and reviewer decision. A reviewer may decide that a word needs a pronunciation rule, a rewritten sentence, or no change at all. Make the decision on a short generated sample, not from the spelling alone.

Vois includes a pronunciation dictionary for approved custom rules. Use it after the language review, not as a replacement for it. The order matters: a dictionary can preserve a chosen pronunciation, but it cannot tell you whether the chosen pronunciation fits the language or setting.

If your reviewer maintains an approved CSV, keep the word,pronunciation,context record with the lesson assets. Add a term to the Vois pronunciation dictionary only after the reviewer approves the spoken form.

Prompt your agent: "Add only the pronunciation entries marked approved in approved-pronunciations.csv to this Vois project. Show the proposed entries and their scope first. Do not add unreviewed terms or change existing global rules."

1. Compare each proposed entry with the reviewer decision and surrounding sentence.
2. Choose project scope unless the reviewer has explicitly approved a term for reuse across lessons.
3. Generate a short sample with every changed term.
4. Have the reviewer approve that sample before applying the entries to a complete series.

The dictionary is a production record, not a way to guess how a new word should sound.

Pace dialogue for replay

Language-learning audio benefits from turns a listener can replay without losing the scene. Keep one intent per turn. Put practical information where a learner can hear it twice. Use punctuation and paragraph breaks as part of the script, not as cleanup after generation.

Compare these two versions:

[Empleado]: Sale a las diez y cuarto, desde el andén cuatro.

[Empleado]: Sale a las diez y cuarto.

Desde el andén cuatro.

The first version is closer to a natural answer. The second gives an instructor a deliberate pause before the platform detail. Neither is universally better. Choose the pacing based on the task: first-listen comprehension may favor the first; a focused repeat exercise may favor the second.

The Vois dialogue-writing skill recommends short, spoken-first sentences, speaker tags only when the speaker changes, and paragraph breaks for meaningful pauses. Read the text aloud with the reviewer before generation. If the writer trips over it, the learner may too.

Turn one clip into several learner exercises

The audio should not be the last thing a learner does. Use the same short scene for a sequence of small tasks that change what they listen for.

1. First listen: Ask a single situation question, such as "Where does the passenger want to travel?"
2. Detail listen: Give two or three choices for the time and platform, then ask learners to select the details they hear.
3. Transcript task: Show the dialogue with the departure time and platform removed. Let learners fill the gaps after another listen.
4. Role switch: Give the learner a new destination or a later departure time. They answer as the passenger or station agent using the dialogue pattern.

Keep the transcript and activity labels separate from the text that goes to the voice engine. The audio script should remain clean, with only speaker labels and the language intended for listening. Your learning platform can add translations, glosses, or answer keys around the clip without making those instructions part of the spoken scene.

For a course with long-form instructor narration, see the e-learning producer's toolkit. This workflow is narrower: short, repeatable conversation clips that teach learners how an exchange unfolds.

A final production check

Before you export a dialogue, confirm these choices together:

• The target-language text has been reviewed for the intended learner range and setting.
• Every recurring role maps to a saved, exact voice ID.
• The selected engine matches the target language and plan availability.
• Names, numbers, and special terms have been heard in a short test clip.
• Paragraph breaks and turn length match the listening task.
• The exported clip has a transcript and an activity that gives learners a reason to listen again.

The goal is not to make voices carry a lesson alone. It is to make a carefully reviewed exchange easy to hear, replay, and use.

Sources

• ACTFL World-Readiness Standards for Learning Languages
• Council of Europe: Common European Framework of Reference for Languages
• Vois CLI skill: project, speaker, generation, and export commands
• Vois dialogue-writing skill: speaker labels and pacing guidance

Give each dialogue a real purpose, then give learners another chance to hear it. Get started with Vois when the script is reviewed, and use multilingual voices to test the target language before producing the series.

The Vois Team