Your Students Can Tell It's You: Record Multi-Speaker Listening Dialogues Alone With Voice Remix (2026 Guide for Language Teachers)

1. Two problems with home-made listening audio

Problem one: every speaker is you

Speaker identity matters in listening tests. Questions like "Who wants to change the booking?" or "What is the woman's attitude?" only work if students have to tell speakers apart by listening. When every part is the teacher, students recognise your voice and use cues you didn't intend to give them, like the slightly silly voice you always use for the second speaker. Exam listening sections use several distinct voices, and IELTS says its listening test uses a range of accents. Students need practice with that.

Problem two: the speech is too clean

Researchers have been pointing at this for years. Elvis Wagner's work on second-language listening shows how unscripted speech differs from the scripted texts learners usually hear in class and on tests. Studies comparing natural and scripted dialogue find significant differences in hesitations, self-repairs, turn-taking and overlap. Learners trained only on clean audio struggle when real people say "so, um, I was going to — actually, no, let's do Thursday".

Text-to-speech dialogue tools fix the first problem and make the second worse. They produce distinct voices reading lines one after another, with perfect timing and no overlaps.

2. Your three options compared

TTS is still the right tool for some jobs: clean beginner dialogues, vocabulary drills and quick model readings. Use the performed-and-remixed method when you want audio that sounds like real people talking.

3. The method: perform, then remix

1. Write the dialogue with natural features marked (section 4).
2. Record each speaker's part as a separate take, performed in your own voice with the hesitations, speed and attitude that character needs.
3. Remix each take into a different target voice with SpeakSay Voice Remix. Voice conversion keeps your timing, pauses, fillers and intonation and changes who it sounds like.
4. Assemble the takes in a free editor such as Audacity, overlapping where speakers interrupt.
5. Normalise the loudness so no speaker is noticeably louder. The free SpeakSay audio normalizer does this in one step.
6. Listen once without the script. If you can't tell who's speaking, neither can your students. Re-cast before you finish.

4. Writing a script that sounds unscripted

A script that reads well on paper usually sounds stiff out loud. Write in the features of real speech and mark them so you remember to perform them:

One hint: real exam distractors often depend on a speaker correcting themselves ("the meeting's at three — sorry, half past three"). Performed speech handles those corrections much more naturally than TTS does.

5. Casting your speakers

Choose target voices that differ in more than one way, so students can tell them apart even through a classroom speaker:

Two practical notes. Remixing across gender or age works best when your performance also moves in that direction with tempo, energy and phrasing; our guides to making a voice sound older or younger and natural-sounding gender conversion explain why. Accent is different. Voice Remix changes who is speaking but mostly keeps how you pronounce things, so your accent comes through. If you need a range of real accents, mix in TTS voices with those accents for the cleaner levels, or invite colleagues who have those accents to record parts, with their consent.

6. Recording turn-taking and overlap alone

The hard part of recording a conversation alone is the timing between speakers. Two methods work:

Method A: line by line, with gaps

Record speaker A's whole part, leaving silence where B speaks. Then play A's track in headphones and record B's part into the gaps, reacting to what you hear. This is the easiest method and gives natural reaction timing.

Method B: overlap in the edit

For interruptions, record B's interrupting line as its own short clip, then slide it in the editor so it starts a quarter of a second before A finishes. Keep A's cut-off word in, since a real interruption stops someone mid-word.

Whichever method you use:

• Record every take in the same quiet room with the same microphone, so the remixed voices sound like they're in the same place.
• Keep takes short, one speaker's part for one scene at a time, so mistakes are easy to redo.
• Remix each speaker's takes with the same target voice and settings so a speaker sounds the same all the way through.
• If you want realistic background noise for higher levels, such as a café or a station, add it quietly at the end, after remixing, across the whole mix.

7. A difficulty ladder from clean to real

Students shouldn't jump straight to messy overlapping talk. Build a ladder and move classes up it:

For independent practice outside class, students can also turn articles into listening practice. Our comprehensible-input guide covers the learner side. For speaking practice, the golden-speaker shadowing method pairs well with this one.

8. Questions that use distinct speakers

Now that each speaker sounds different, write questions that depend on it:

• Attribution: "Who suggests changing the date?"
• Attitude: "How does the daughter feel about the hotel?"
• Agreement: "What do both speakers agree on?"
• Corrections: "What time does the man finally choose?"
• Interruption: "Why does the receptionist stop mid-sentence?"

These questions don't work with a one-voice recording, and they're what students meet in real exams.

9. Ground rules for the classroom

• Use only your own performance. Don't remix students' voices or colleagues' recordings without clear permission, and never with children's voices.
• Don't imitate real people. Use generic target voices, not celebrities, politicians or anyone students would recognise.
• Tell students how the audio was made. A one-line note ("the teacher performed all parts; the voices were changed with AI") is honest and often interests them.
• Follow your exam board's rules for anything used in formal assessment. For class practice and internal tests, the method above is fine.
• Keep a master copy of your takes. If a remix doesn't work, you can re-cast without re-recording.

10. FAQ

Record each speaker's part as a separate take in your own voice, then use voice conversion such as SpeakSay Voice Remix to turn each part into a different target voice. Assemble the takes in an audio editor.

TTS gives you distinct voices, but its speech is clean and turn-by-turn, without the hesitations, self-repairs and overlaps of real conversation. It's good for beginners. Performed and remixed audio is better for realistic practice.

Mostly no. It changes the voice's identity, such as age, gender and timbre, while largely keeping your pronunciation and rhythm. For a range of accents, mix in TTS voices with those accents or recordings from consenting colleagues.

Expect about 45 minutes for the first one, including writing, recording, remixing and assembling. With a template and fixed voice settings, later ones are much quicker.

Follow your exam board's rules for formal assessments. For classroom practice, homework and internal tests, it's a practical way to make realistic multi-speaker audio.

Yes. A short note is honest, avoids confusion and often becomes a useful conversation about how real speech works.