We make OmniTalk, a two-way live voice translation app for iPhone. This article is not about picking apps. It is about what to do with your mouth, alone. If you want specific tool comparisons, read the tools roundup instead.
Five methods follow. Four are preparation. The fifth involves another human, and it is the only one that actually moves you from rehearsing to speaking. The other four make that fifth conversation less intimidating and more useful.
Method 1: Shadowing — imitate audio in real time
Shadowing is repeating recorded speech aloud while you hear it. It trains prosody, connected speech, and the physical act of making target-language sounds.
This week:
- Pick 30–60 seconds of audio you have a transcript for. A podcast dialogue, news segment, or textbook CD track works.
- Play it once while reading the transcript.
- Play it again and repeat aloud, staying about one or two syllables behind. Do not pause.
- Do that three times. Record the fourth attempt.
- Listen back with the original. Mark where your stress fell differently.
Ten minutes a day is enough. If it falls apart, slow the audio to 0.75x, not 0.5x. Very slow audio destroys stress patterns. If it feels easy, use a news report rather than a slow learner track.
You will sound clumsy for the first three days. That is not a comprehension problem. It is a motor problem. The mouth has not learned the movements.
Shadowing does not ask you to create sentences. It will not correct you. It prepares the mouth and ear. That is enough for one method.
Method 2: Self-talk narration — speak your own life aloud
Narrate your own actions aloud as you do them. This trains retrieval of words you actually need and assembly of simple sentences without a listener.
This week:
- Choose a five-minute daily slot: washing up, making coffee, walking to the station.
- Describe what you are doing. "I am opening the fridge. I need eggs. The milk is on the middle shelf."
- After two days, switch to past tense. "I opened the fridge. I needed eggs."
- After another two, switch to future. "I will open the fridge. I will take the eggs."
If you do not know a word, write it down and look it up after the five minutes. Interrupting to check kills the flow. If you have OmniTalk, you could point the camera at an object to get the word later, but only after the slot.
The main limit is that no one corrects you. You can fossilise an error if you never check. That is why Method 3 exists.
Method 3: Recording and listening back — hear your own foreign voice
Record yourself speaking, then listen. This is the least comfortable solo method and one of the most useful.
This week:
- Choose a simple prompt in the target language: "What did you do yesterday?" or "Describe your flat."
- Record a two-minute answer. No notes, no re-recording.
- Wait until the next day. Then listen at normal speed.
- Write down three things that were wrong or unclear. Check one grammar point.
- Re-record the same answer once.
Do this three times this week, about 15 minutes each session. You will dislike the sound of your voice. Do it anyway. The value is hearing what you actually said, not what you intended. A saved transcript can help if your recorder produces one, but a plain voice memo is fine.
Do not re-record until you have a clean take. The first take is the data. If you re-record until it is smooth, you are practising performance, not language.
Recording and listening back trains noticing. It does not give you interaction. But it catches errors that self-talk hides.
Method 4: Scripted rehearsal of the twenty exchanges you will actually have
Most speaking practice is too broad. "Ordering in a restaurant" is useless if you will not eat out this week. Replace it with the twenty exchanges you are most likely to have.
This week:
- Write the real situations on your calendar. Check into a hotel. Ask where the lift is. Tell a taxi driver your street. Say you do not eat shellfish. Ask for a receipt.
- Write each exchange as two to four short turns. Not paragraphs. "Do you have a room? / Yes. / One night. / How much?"
- Rehearse aloud until the phrases come without looking.
- Then change one variable: the room is too small; the taxi does not accept cards. Rehearse the new turn.
Write the exact turns, not topic headings. "At a museum" is a topic. "Does this ticket include the special exhibition?" is a turn.
Use a phrasebook if you need starting turns. OmniTalk has printable travel phrasebooks at /phrasebook, but any phrasebook works.
Scripted rehearsal trains high-frequency chunks and buys working memory. The limit is obvious: real people deviate from your script. But the script is not the conversation. It is a warm-up. When a real person deviates, you still have the chunks.
Method 5: Talk to a real person with a translation app as a safety net
This is the only method that involves another human. It is the one that actually moves you. The previous four methods are preparation for it.
This week:
- Find one person: a neighbour, shopkeeper, conversation exchange partner, or a paid tutor if you can afford it.
- Set a phone between you. Use a two-way live voice translation app, OmniTalk or something else. We make OmniTalk, so note the bias.
- Speak in the target language first. Do not rehearse. The point is to survive a real exchange.
- If it collapses, let the app translate. Read the saved transcript afterwards.
- Keep it to 15 minutes. One live conversation a week is enough to start.
If you use OmniTalk, the free tier is one hour of live conversation per month. PRO is unlimited. Camera and text translation do not count against that hour. It works offline once you download a language pack. BASIC uses Apple's on-device speech recognition and translation; PRO uses a premium cloud speech engine. That is specification, not a recommendation.
A paid human tutor beats an app for corrections and for adapting to your level. Use one if you can. A translation app simply lowers the cost of a broken sentence enough that you will open your mouth.
This method is the only one that involves another human, and it is the only one that moves you. If you only do one thing, do this one.
How the methods compare
| Method | What it trains | Time this week | Main gap |
|---|---|---|---|
| 1. Shadowing | Mouth, ear, connected speech | 10 min/day | No sentence creation |
| 2. Self-talk narration | Retrieval, sentence assembly | 5 min/day | No correction |
| 3. Recording and listening back | Noticing, self-monitoring | 15 min × 3 | No interaction |
| 4. Scripted rehearsal | Automatic chunks, readiness | 20 min × 3 | Real people deviate |
| 5. Live conversation with safety net | Actual interaction, repair | 15 min × 1 | Requires another human |
A week of solo work that ends with a real conversation
Use this schedule if you have no partner most days but can find one person once a week.
Monday to Friday: 10 minutes shadowing plus 5 minutes self-talk.
Tuesday, Thursday and Saturday: 15 minutes recording and listening back.
Wednesday and Sunday: 20 minutes scripted rehearsal.
Saturday or Sunday: 15-minute live conversation with a translation app as a safety net.
That totals about 2 hours 55 minutes across the week, excluding travel to the live conversation.
If you only have 30 minutes a day, run shadowing and narration daily, and alternate recording and scripts. But nothing replaces the weekly live session.
If you cannot find a person this week, do methods 1–4. You are not practising speaking; you are rehearsing it. Book the live conversation anyway.