Download this bank as PDF

Listen first — sample clip types

Synthetic (TTS) examples — a guide to the shape of each clip type, not a voice to imitate.

Code-switch
Disfluency
Pause

French recording battery

Thank you for recording these. There are sentences below; each row tells you how to read it — plain ones naturally, and the marked code-switch / disfluency / pause / one-word clips following the note in their row. Listen to the three samples above first to hear what those clip types sound like.

How to record each clip

  1. Find a quiet room, start recording once (Voice Memos or equivalent is fine), and keep the phone recording the whole way through — you don’t stop and start between sentences, it’s all one recording. Read down the list in order.
  2. Leave a clear gap of about five seconds of silence between one sentence and the next — just wait quietly, then start the next one. Those gaps are how the single recording gets split back into individual clips afterwards, so make them clearly longer than any pause inside a sentence. If you fumble a sentence, just pause, say the whole sentence again, and carry on — the last clean take of each is what’s used.
  3. Read each sentence naturally, at your own pace. If you’re not fluent in the language, that’s fine — read it as a learner would: slow down, mispronounce things, self-correct. That IS the population this benchmark measures. Don’t over-produce it — hesitation and imperfection are the data, not noise to eliminate.
  4. For clips marked code-switch: say the [[bracketed]] word or phrase in English, out loud, right in the middle of the sentence — like you genuinely forgot or never learned the word in the target language. Don’t translate it.
  5. For the disfluency clips: follow their instruction (read through an inserted filler word naturally, cut yourself off at the em-dash and restart the word right after it, read slowly and deliberately, or pause partway through). The short pause inside one of these sentences is part of the sentence — keep it shorter than the five-second gap you leave between sentences.
  6. For the robustness probe clips: follow them exactly — either say only the single word and nothing else, or (for the silent one) stay silent for a few seconds. These exist to catch a model hallucinating words that were never said, so it matters that nothing extra gets said — then leave your normal between-sentence gap and move on to the next.

The battery — all 45 clips

# Type Say Instruction
1 disfluency (pause) Ma nièce joue du piano. Read this sentence, but pause for a few seconds partway through, like you’re thinking mid-sentence, then finish it.
2 disfluency (slow) Il a un chat noir. Read this slowly and deliberately, like you’re carefully working out each word as a learner would.
3 disfluency Il ben porte un tee-shirt orange. Read this naturally, including the filler word exactly as written, like a real hesitation.
4 disfluency Cette mai— maison heu est très vieille. Read this naturally, including both the filler word and the self-correction exactly as written — the em-dash (—) marks where you cut yourself off and restart the word right after it.
5 disfluency Oui, mer— merci, c’était très bon. Read this naturally, including the self-correction exactly as written — the em-dash (—) marks where you cut yourself off and restart the word right after it.
6 disfluency Un thé ben au lait, s’il vo— vous plaît. Read this naturally, including both the filler word and the self-correction exactly as written — the em-dash (—) marks where you cut yourself off and restart the word right after it.
7 disfluency C’est euh loin d’ici ? Read this naturally, including the filler word exactly as written, like a real hesitation.
8 disfluency Le serv— serveur heu apporte les menus à la table. Read this naturally, including both the filler word and the self-correction exactly as written — the em-dash (—) marks where you cut yourself off and restart the word right after it.
9 disfluency Je me perds da— dans la ville. Read this naturally, including the self-correction exactly as written — the em-dash (—) marks where you cut yourself off and restart the word right after it.
10 disfluency Ce restaurant heu est trop cher pour moi. Read this naturally, including the filler word exactly as written, like a real hesitation.
11 disfluency Oui, un ki— kilo de pommes, s’il vous plaît. Read this naturally, including the self-correction exactly as written — the em-dash (—) marks where you cut yourself off and restart the word right after it.
12 disfluency (pause) Ce journal est considéré comme une source fiable. Read this sentence, but pause for a few seconds partway through, like you’re thinking mid-sentence, then finish it.
13 disfluency Le ben racisme est un fléau qu’il faut comba— combattre par l’éducation. Read this naturally, including both the filler word and the self-correction exactly as written — the em-dash (—) marks where you cut yourself off and restart the word right after it.
14 disfluency Il heu faut boire au moins un litre d’eau par jour. Read this naturally, including the filler word exactly as written, like a real hesitation.
15 disfluency La ben mémoire joue un rôle fondamental dans l’apprentissage. Read this naturally, including the filler word exactly as written, like a real hesitation.
16 disfluency Le raisonnement ben scientifique repose sur des faits vérifiables. Read this naturally, including the filler word exactly as written, like a real hesitation.
17 disfluency Tu ben les lui montres. Read this naturally, including the filler word exactly as written, like a real hesitation.
18 disfluency J’aime ben le raisin blanc. Read this naturally, including the filler word exactly as written, like a real hesitation.
19 disfluency (slow) Mon père adore le jardinage, il a un beau potager. Read this slowly and deliberately, like you’re carefully working out each word as a learner would.
20 disfluency (pause) Tout le monde souhaite vivre en paix. Read this sentence, but pause for a few seconds partway through, like you’re thinking mid-sentence, then finish it.
21 disfluency La euh récolte des cerises a lieu en juin dans le sud de la France. Read this naturally, including the filler word exactly as written, like a real hesitation.
22 disfluency Je ben suis en train de manger. Read this naturally, including the filler word exactly as written, like a real hesitation.
23 disfluency (pause) Ils n’ont pas demandé de frais pour cette transaction. Read this sentence, but pause for a few seconds partway through, like you’re thinking mid-sentence, then finish it.
24 disfluency Ne sois pas égoïste, partage av— avec ton frère. Read this naturally, including the self-correction exactly as written — the em-dash (—) marks where you cut yourself off and restart the word right after it.
25 plain Les employés de l’entreprise ont reçu une prime. Read this sentence naturally.
26 plain La projection est un mécanisme de défense où l’on attribue ses propres émotions à autrui. Read this sentence naturally.
27 code-switch (noun) je dois prendre un heu [[discount]] demain matin Read the sentence, but say the [[bracketed]] word in English, like you forgot the French word for it.
28 code-switch (noun) j’ai oublié mon [[schedule]] à la maison ce matin Read the sentence, but say the [[bracketed]] word in English, like you forgot the French word for it.
29 code-switch (noun) il me faut un [[receipt]] avant la réunion de demain Read the sentence, but say the [[bracketed]] word in English, like you forgot the French word for it.
30 code-switch (noun) on a un [[appointment]] la semaine prochaine je crois Read the sentence, but say the [[bracketed]] word in English, like you forgot the French word for it.
31 code-switch (noun) il me faut un [[grocery store]] avant la réunion de demain Read the sentence, but say the [[bracketed]] word in English, like you forgot the French word for it.
32 code-switch (noun) on a un [[road trip]] la semaine prochaine je crois Read the sentence, but say the [[bracketed]] word in English, like you forgot the French word for it.
33 code-switch (how-do-you-say) ben comment dit-on [[appointment]] en français ? Read the sentence, but say the [[bracketed]] word in English, like you’re asking how to say it.
34 code-switch (how-do-you-say) comment on dit [[deadline]] déjà ? Read the sentence, but say the [[bracketed]] word in English, like you’re asking how to say it.
35 code-switch (how-do-you-say) je ne sais pas comment dire [[schedule]] en français Read the sentence, but say the [[bracketed]] word in English, like you’re asking how to say it.
36 code-switch (how-do-you-say) ben comment dit-on [[insurance]] en français ? Read the sentence, but say the [[bracketed]] word in English, like you’re asking how to say it.
37 code-switch (clause) c’est un peu ben [[how do I explain this]] compliqué à dire Read the sentence, but say the [[bracketed]] phrase in English, like you switched languages mid-thought.
38 code-switch (clause) hier je suis allé au marché et euh [[wait I forgot the word]] Read the sentence, but say the [[bracketed]] phrase in English, like you switched languages mid-thought.
39 code-switch (clause) elle m’a demandé si ben [[sorry let me think for a second]] Read the sentence, but say the [[bracketed]] phrase in English, like you switched languages mid-thought.
40 robustness probe (one-word) oui. Say only this one word, then stop — nothing else before or after.
41 robustness probe (one-word) oui. Say only this one word, then stop — nothing else before or after.
42 disfluency (pause) Quelle est votre taille ? Je fais du trente-huit. Read this sentence, but pause for about 5 seconds partway through, like you’re thinking mid-sentence, then finish it.
43 disfluency (pause) Elle a terminé son travail avant la date limite. Read this sentence, but pause for about 5 seconds partway through, like you’re thinking mid-sentence, then finish it.
44 robustness probe (trailing silence) Le quinze août, beaucoup de Français sont en vacances. Read this sentence normally, then stay silent for a few seconds afterward before stopping, like you trailed off without saying anything more.
45 robustness probe (trailing silence) Sa femme est médecin. Read this sentence normally, then stay silent for a few seconds afterward before stopping, like you trailed off without saying anything more.