AI may create lots of pronunciation mistakes specific for French voiceover, all of those need to be fixed.

1. Pronouncing a silent letter at the end of a word: E.g., “es” is fully pronounced, “pars” or “veux” as well.

2. Pronouncing silent verb endings. E.g., “disent” may be read as spelled or “dizt”.

3. Sound omission: “Dan m’a dit” becomes “Dan a dit”; “surpris” is pronounced as “surris”.

4. Lack of liaison or it’s appearance where it’s not required. E.g., “Tu veux une preuve ?” can be pronounced with liaison between the verb and the article.

5. Mispronunciation: “lundi” is pronounced as “loundi”.

6. Appearance of random sounds in random places - extra sounds or just butchered existened sounds. This may present itself is any word, any sound, and is absolutely random: “Prévent” (Presént), “saire” (faire), “sille” (fille). More frequent “butchering” are: “nn” instead of “ll” (“elle” -> “enne”), “s” instead of “f” (“fille” -> “sille”), “v” instead of “z”-sound (“Ils sont” -> “il vont”). Basically, any sound (mostly consonants) may be butchered like this.

1. Revoice - there’s a slight chance that a simple revoicing will be enough to fix it. Especially with i. 3 and 6. Liaison is sometimes random, so revoicing a few times may help to make it right (present or absent depending on your needs).

2. Changing spelling. You can change spelling based on your needs and disregarding the correct spelling and/or grammar: if you end up with a non-existent word or a false grammar structure, it’s nothing to worry about - the most important thing is that the voiceover sounds as it should.

E.g., one of the most frequest issues is pronouncing the “-ent” ending in verbs. What you can do, is eliminate the ending all together, since it’s silent. But make sure to spell the word in a way that sounds the same as its pronunciation with the ending: disent -> dise (don’t write “dis”, as it will either be pronounced “dis” and not “diz”, or even “di”), dorment -> dorme, etc.

In one movie, “beauté“ was read half-English and half-French, with the stem read as in English “beauty”. The simplest solution was to remove “e” from the stem, so the remaining letters created a fully French (in terms of phonetics) word: “bauté”. This worked every time.

Someone was also able to find a smart solution to spell “Qu’est-ce que” - a lot of the times it is pronounced incorrectly or is stretched too much (more on this belowe). Alternative spelling is “Kesque”, AI almost never fails to pronounce it right.

Verbs “es”, “est” are often mispronounced, so it’s better to spell them as “é“ if that’s the case. So, simplest “Qu’est-ce que c’est” may become “Kesque cé” or “Kesque c’é”. As terrifying as it might look, it works well for AI voiceover.

You can also omit endings in words like “fait”, “fais” (basically, any silent endings not participating in liaison). There’s a chance that AI will read “fai” just like it’s spelled, and not “fé”, so you can try respelling it again (e.g. as I did here). Different „é“ sounds are not going to be as perfectly correct as when the words are spelled normally, but it’s an option to get a much better result than that without changing the spelling.

All of the above are not strict rules, they’re guidelines that you can follow to get better results and that can give you ideas and inspire you to try other alternative spellings that will give you the required result.

A lot of the time, especially when it comes to shorter blocks, AI doesn’t understand that it has to speak French. As some words or even phrases are etymologically related to those in other (mostly Roman) languages, AI often mistakes them for those of other languages. Pay attention to sounds (e.g. “R” in French and Spanish sounds very different), intonations (sometimes you can hear Italian intonation, which may sound quite stereotypical), and overall pronunciation of words. The example with “beauté” arose because of this same issue: as English derives lots of words from French, AI couldn’t decide whether to read the word in English or in French.

More examples: “six” is read in English, “merde” is read in Spanish, “assistante” is read in Portuguese, “fête”is read in Italian.

Sometimes it’s not even clear what language AI reads it in - probably, no one in particular, it just reads all the letters in the word/words. E.g. “t’es vue” is read as “tesse voué”.

Solution: Add some French text at the beginning (before the main text) - it should be one sentence 3-6 words long. Normally, it helps AI understand that it’s French and should be pronounced as French. If there are still problems, you can try adding more text - but now after the main one (so you have some text before AND after the main one), changing spelling, or rephrasing. If nothing else helps, the remaining option is just to change the challenging words to synonymous ones.

Often, French translation differs in length from the original text. Here’s what you can do:

1. Lengthy blocks 1. Sometimes short blocks (containing 1-2 words) are stretched in the voiceover so now they become too long: e.g. words like “mais”, “(eh) bien”, “alors”, “et” “d’accord”, etc. Say, we have “but” in the original, but the translation goes like this: “maaaaaaais”. Basically, AI may do this to almost all words if they’re single-standing in a separate block.

In the example below, the original track is cut short a little bit (like, you can hear the “O”, but it doesn’t go all the way). But in the translation, “Bonjour” is quite slow. Not necessarily “Bonjouuuur”, but still quite long, as you can see in comparison (of the waves). “Bonjour” in itself doesn’t have to be much longer than “Hello” (say them both and compare, they’re very close in length), so we may be able to achieve a shorter audio. The yellow arrow shows that the voiceover track can be scrolled to the right - that’s how long it is.

Possible solutions:

1) Replace “Bonjour” with a shorter version (e.g. Salut, but this word can also be stretched + pay a lot of attention to the pronunciation - AI tends to mispronounce the last letter in this word).

2) If you have “…” or absolutley no punctuation at the end of the text, put a period there - “…” and lack of punctuation often cause the voiceover to be more stretched or have funny intonations. A period shows a lowering of intonation characteristic of human speech. When AI doesn’t do that, often it sounds as just that - AI, which is not what we need.

3) Sometimes simple revoicing will help (even though it’s rarely the case). If you have a “?” or an “!” with a space before them, try removing it.

4) Add some text afterwords, a sentence of 3-5 words. In this case, AI kind of has a “limit” in time and space, as there’s another sentence to say, and it can’t just say “maaaaaaais” (for example) indefinitely. You may need a few attempts to get a better result, and this doesn’t work perfectly or a 100%. Sometimes you’ll do it first try, sometimes you’ll only be able to make the audio slightly shorter. But it’s already better than nothing. Don’t forget to cut the extra piece out in Audacity.

1) Here’s an example of how removing a space before a “?” helped shorten the block (the wave is scrolled a little due to the blank space in front of the text, which we don’t count):

2) Here changing punctuation helped, too. First version was “Ouais…?”, and I changed it to “Ouais?”. As simple as that. So check punctuation first.

3) This is how adding some text works. It was impossible to get a shorter “que” in this block, so the next step is to add some text. It helped in a few attempts, you can see how short this wave is, while the middle one is very long. The rest of the length in the bottom wave is this extra phrase I added. Now save this block, download it and cut out the extra sentence in Audacity. Don’t forget to upload the new audio and to delete extra text. This text might go to people who don’t speak French, so they won’t recognize that this text is not needed and revoice the whole thing.

2. Lengthy blocks 2. Sometimes blocks may be too long due to some lengthy structures and phrases. This especially goes for “Est-ce que”/ “Qu’est-ce que”, and similar variations.

First, AI struggles with reading this, and a lot of the time it doesn’t pronounce the structure in a concise colloquial manner, but in a more stretched one - not “esse que” (just a transcription), but “esse ce que”. And sometimes it even adds an extra “que” to the equation, and we get something like “esse ce que que”. Second, consequently, AI may pronounce this structure for what seems like an eternity. And then we end up with the block “What?” having a voiceover of “Qu’est-ce que que c’est ?” that lasts three times longer than the original. This structure is not banned or anything, but it’s problematic, so if you see AI getting stuck in it, consider rephrasing. E.g. if the conversation is not too formal in tone, just replace it with either “C’est quoi ?” or “Quoi ?” (if context allows).

3. Lengthy blocks 3. Main strategies to shorten a sentence: 1). Rephrase: Find synonymous shorter words or structures. The “Magic Wand” button may help you with this, it’s located right under the audio wave to the right. At least, it may give you an idea of what you can do. 2). Eliminate redundancy. E.g.: “Je tiens un livre dans mes mains” - of course you’re holding it in your hands, and not feet (for example), meaning you can remove “dans mes mains” without losing anything in terms of meaning. Same goes for so-called parenthetical and modal words, i.e. those that may add some kind of emphasis and connotations/modality but they don’t change the meaning of the sentence: “En fait”, “Donc”, “Alors”, “Très“, “Tellement”, etc. And you can also remove all redundant information (like the one about the hands): “Pourquoi tu l’as fait ?” - if context allows it, you can only write “Pourquoi ?”, if from the scene it’s clear what this question means. Same goes for “Marie! Est-ce que tu es là ?” -> “Marie !” or “Tu es là ?”. It strongly depends on the context, but more often than not, a lot of stuff may be eliminated. However, don’t overdo it! Check if the sentence is still grammatically correct and clear.

E.g. sometimes, if AI has nothing else to offer, the “Magic Wand” will give you weird ideas. Don’t follow them blindedly. It may offer changing “Je tiens un livre” to “Un livre”. In this case, there’s no sense in this sentence (very rarely it may fit the context, then it’s okay), so don’t fall for it, AI’s often mistaken. Be logical and apply critical thinking to what you do.

1) Here I changed “Je pense…” to “Alors”. The whole block is basically a filler. There’s no “Je pense que” in the next block - she just says a different sentence (that’s what she thinks, but it’s not necessarily a continuation of this phrase). Therefore, this change won’t harm the meaning and the structure of the next block.

2) Here I changed “Ça va être…” to “C’est ça”, which is appropriate in this context.

Context: The speaker says “It’s not gonna be easy”, then this phrase - “It’s gonna be…”, then “It’s not gonna be easy” again. In this case it’s just repeating the same thing, so it’s appropriate to rephrase a little, like in the previous example. It won’t harm the meaning at all. I tried a few options like “ça ne…” and “Je veux dire”, but they were longer, so I stuck to this one.

1) The original: You looked like black Abraham Lincoln the last time.

Translation: Avant tu étais comme Abraham Lincoln en noir.

Updated translation: Tu étais comme Abraham Lincoln en noir.

In the context and with the past tense used we may get rid of the specification “avant”. The audio is much closer in length now.

2) The original: I don't think I went too overboard this time around.

Translation: Je ne pense pas avoir fait trop d'excès cette fois-ci.

Updated translation: Je ne pense pas avoir fait trop d'excès.

We can get rid of “cette fois-ici” because there’s certain context in this conversation (it’s the same one as above), and both speakers know what time they’re talking about. It’s worth noting that I slowed this audio: it still sounds okay, not too slow, and the version without the slowing was shorter. But you can check if a shorter version creates a red block, and if not, just let it be and not slow the audio down.

4. Short blocks 1. Shorter blocks may appear because of the original speaker shouting our just prolonging some sounds. We may try to fix it by either adding some more vowels or by toggling the “Slow Audio” option.

1) Adding vowels is a bit tricky for French: sometimes AI just stops recognizing words as French anymore. A good example would be, “Non” -> “Nooon”. The second one loses its nasal sound while revoiced. One solution is to try replacing a regular vowel with the one with an accent circonflexe. Other diacritics weren’t convincing enough, so AI didn’t perceive them as French. But “ô” helps prolong the vowel while still maintaining the French pronunciation. So, “Non” -> “Nôn”. Otherwise, try different things to get a better result, it’s all about creativity and critical thinking here.

2) Slow Audio: be extremely careful here! This option may give you a normal-paced but a little slower result, which is great. But it also may generate a very slowed-up version that you will recognize as slowed if you hear it in a movie (or anywhere). Be extra vigilant about this. So, if you have 1-2 words being stretched in the original (“yeeeeeeah”, “Moooooom!”, etc.), you may try using this option).

Example: Here I used both adding fillers (see info below) and I slowed the audio. The pace is appropriate, and I added another “juste” for length, which doesn’t sound too reduntant and matches the way people talk sometimes.

Original translation: Et c’est juste… ça se termine trop tôt.

Updated translation: Et c’est juste… ça se termine juste trop tôt.

5. Short blocks 1. To lengthen other sentences, most of the time you’ll need to use rephrasing again. Basically, it’s the process similar to shortening, but in reverse: we either find longer synonymous structures/words, or rephrase to create redundancy. First, you start with looking for synonyms, and if you can’t, then you can start adding some fillers or words that don’t affect the initial meaning. Don’t overuse it, constantly check if the sentence still sounds authentic. The same things you eliminate while shortening, you can add for lengthening: En fait, Alors, Bien, Bon, Ouais, etc. Also you can add some filler phrases like: Je sais, Tu sais, Je crois, c’est ça, je veux dire, or basically anything that will be adequate in the conversation and not affect the meaning. Let’s go back to the example with “Mary!”, now we need to lengthen it. Depending on the context, we may add things like “Mary ! C’est toi !” / “Mary! Où etais-tu ?”, a more obvious one would be “Mary ! Mon Dieu !”.

1) As mentioned, you can use “Magic Wand” for ideas. Here’s one example:

Revising this translation we have to stay attentive and check for authenticity and for length. Here it’s obvious that there are way too many words added (also compare to the length difference that you have in the block (the percentage), but we may get inspired from the idea and use “vraiment” and/or “très” only.

2) Here the original is much longer, and the text is actually much longer, too, so some words are missing here.

Original text: And alongside that, regardless of...

Original translation: Peu importe...

Updated translation: En fait, pue importe é... (I also slightly changed the spelling due to mispronunciation of “est”. The audio now is twice as long and fits much better in terms of length.

1) In this sentence, you can change “parfois” to “de temps en temps”, and/or change “penser” with “considérer” or “avoir l‘impression”. Just look for other ways you could say the same sentence.

2) Here I did both rephrasing and a spelling correction.

Context: Phrases before are - “Is she still looking down on us? Isn’t she?” and then this reiteration “Isn’t she?”.

Original translation: “Oui?”.

As you can see, it’s too short and is marked as sufficient due to the blank space before and after. It’s not okay and we have to change these blocks to make the text length closer to the original. In this case, percentage is not really telling.

What I did is rephrasing “Oui?” to the approprate in this context “N’est-ce pas?”. The problem was, it was too long due to a not concise pronunciation. So I changed the spelling to make the pace more concise. And it helped. + due to the new spelling, AI had no chance of butchering the pronunciation.

! Each name in the movie should have one and the same pronunciation throughout the whole film !

First choose one version of a name for each character and place (or main ones, if there are too many of them), create a list and write all of them down with stresses and comments about other pronunciation specifics. Send this list to your editor when you submit the project. LUcy or LucY? JImmy, JimmY, or may be “J” will be French here, and not English? AlExa or AlexA? In the list, write exactly how you want to call a character or a place - e.g. “JImmy with an English “J”, or “JimmY with a French “J”. Thus, you’ll have something to stick to (and not just your memory which is not always helpful when you do a big project), and your editor will know what to look out for, and not ask you every time which version of the name you actually chose.

It’s important to note that with French, there are two challenging things: stress and overall pronunciation. You can pronounce the name “Jasmine” as “DjAsmine”, DjasmIne”, “JAsmine” or “JasmIne” (where “J” is French). Sometimes it may even be read with German “J”, so “YAsmine/YasmIne”. AI is multi-lingual, which creates an issue here, especially when a name stands alone in a block.

NAMES LIST FORMATTING

When handing a project over, please send a list of proper names mentioned in the movie to your manager - both characters’ names, and geographical names. You can send it together with other notes and comments you want to share. You can send this list either as a simple text or as a Microsoft Word document.

Formatting

The list should include: 1) Proper name as used in the original. 2) Description of pronunciation that you use in the voiceover. We need to know whether a name has English of French pronunciation and which syllable is stressed. So, it’s not just a list of names you saw in the movie, it’s a guide on how each proper name is pronouned in your voiceover, and your editor should be able to know this by looking at your list.

Example:

Mary - pronounced as an English name, Y is stressed.

James - “J” is pronounced in French, other sounds are English.

Lola - LolA, French stress.

Andrew - pronounced in French - Andreu, last syllable stressed.

Ramsey - pronounced “Ramzy” with Y stressed.

If needed, you can add some comments to the list, especially if there are real people and places mentioned - you can include this note in the list and let us know that you checked how those real people’s names are pronounced in French (I.e. Dakota, General Sibley, King Charles, Iowa, etc.).

One way to deal with names more efficiently is to use an Excel file of the project.

1. Download the Excel file, you can do that pressing the circled button right under the video:

2. Once you’re done, open the file. That’s what you’ll see. You’ll need the last column, it’s translated as “Translation”. Highlight it by putting the cursor the the letter “G” about the column name, the cursor will become a small arrow, then left-click with the mouse.

3. After that, press Ctrl+F on your keyboard. You’ll see a smaller window. In the entry field, write the name that you need. «Найти» here means “Find” (1). Then press button (2), it means “Find all”.

4. After doing that, that’s what you’ll see. It’s the list of all iterations of the name in the whole translation. By clicking on a row, you’ll be directed to the exact line (block) where this name is used. In the “C” column you can see the number of the block. Thus, you can write down the numbers of all blocks where a certain name is present and check how it’s pronounced. This will allow quicker search and fixing of the pronunciation if needed. It’s not mandatory to do it this way, but just one way to fascilitate the process.

1. For convenience, it’s better to choose a fully English or French version of a name: DzImmy, or JimmY. So, choose the one where all reading rules apply, and both vowels and consonants are pronounced either in the English manner, or the French one. Otherwise, you’ll have even more trouble trying to achieve, say, an English “J” and the French stress (i.e. the last syllable stressed).

2. French version. Oftentimes, with the names spelled the same in English and French (and just pronounced differently), having a name in the midst of a French text will be enough. If not, changing spelling is key here. The same rules apply here as with changing spelling for any other words to be pronounced correctly. For example, you could write “Jimmi”/”Jimi”/”Ji-mi” etc. Maybe even “Zhimi”. If there’s an issue with stress, you can write the stressed syllable in an upper case: “JiMI” - for some reason, it’s less effective with French than with other languages, but it may also help. You can also get some stresses right by using vowels with diacritics. For example, someone really wanted to use a more French version of the name “Michael”, so this is the spelling they came up with: Maikél.

3. English version. If a name is not pronounced in English the right away, the first thing to try would be to put it into quotation marks. This often helps with any foreign speech in general (when someone says a foreign word in a French sentence), but there’s one thing to note: sometimes AI may stop for a moment before saying the word in quotation marks, thus creating an unnecessary pause and making the whole thing sound off. Sometimes this may be fixed by a few revoicings, but sometimes it can’t be fixed at all. You may try rephrase the sentence and changing the position of the name in the sentence (but so that it still makes sense) - this may prevent AI from pausing. You can also try changing the spelling - e.g. you could write in quotes something like “Gymmy/Gymmi” or “Dzhimi”, and/or use uppercase to highlight the stressed syllable.

4. If a name stands alone in the block. The first thing to do would be, as always with such blocks, to add some text before - chose the language depending on what pronunciation of the name you would like to have. If you choose English, check if the name here is pronounced the same as in other blocks: if in other blocks this name is pronounced in English BUT with a French accent, and here it’s pronounced with an English accent, then it’s a problem. Then add French text instead and try other strategies outlined above. Occasionally, changing spelling and punctuation alone may help: if you write “Jimmi !”or “Jimmí !” this may work. Once I had the name “Destiny” alone in a block, and I needed to stress the last syllable. I wrote it as “Déstiny“ or “Déstini”, and it helped, as “é” is not a part of the English alphabet.

1. A block should always have a punctuation mark at the end! It may be a period, three dots “…”, an exclamation point “!”, or a question mark “?”. It will rid all of us of unnecessary fluctuations in intonations. No punctuation mark often causes flat AI-like intonation, which we can’t voice a movie with.

2. Voiceover will sometimes be more adequate in intonation, if we put a space before a «!» or a «?» - some slight issues may be fixed with this. Also, if you have really short blocks, and there is a “!” or a “?” at the end, check if there’s a space there. Often, it will help AI to know that this is French, so there’s a bigger chance it will pronounce the word correctly. You can check this both before the voicing, and after, if a block is mispronounced.

E.g., “Non!” and “Non !” can sometimes be pronounced the same, but oftentimes, AI will pronounce the first one without the nasal sound, and the other one - with it.